Daily Papers

Pre-trained models have revolutionized natural language understanding. However, researchers have found they can encode artifacts undesired in many applications, such as professions correlating with one gender more than another. We explore such gendered correlations as a case study for how to address unintended correlations in pre-trained models. We define metrics and reveal that it is possible for models with similar accuracy to encode correlations at very different rates. We show how measured correlations can be reduced with general-purpose techniques, and highlight the trade offs different strategies have. With these results, we make recommendations for training robust models: (1) carefully evaluate unintended correlations, (2) be mindful of seemingly innocuous configuration differences, and (3) focus on general mitigations.

Collaborative perception, fusing information from multiple agents, can extend perception range so as to improve perception performance. However, temporal asynchrony in real-world environments, caused by communication delays, clock misalignment, or sampling configuration differences, can lead to information mismatches. If this is not well handled, then the collaborative performance is patchy, and what's worse safety accidents may occur. To tackle this challenge, we propose CoDynTrust, an uncertainty-encoded asynchronous fusion perception framework that is robust to the information mismatches caused by temporal asynchrony. CoDynTrust generates dynamic feature trust modulus (DFTM) for each region of interest by modeling aleatoric and epistemic uncertainty as well as selectively suppressing or retaining single-vehicle features, thereby mitigating information mismatches. We then design a multi-scale fusion module to handle multi-scale feature maps processed by DFTM. Compared to existing works that also consider asynchronous collaborative perception, CoDynTrust combats various low-quality information in temporally asynchronous scenarios and allows uncertainty to be propagated to downstream tasks such as planning and control. Experimental results demonstrate that CoDynTrust significantly reduces performance degradation caused by temporal asynchrony across multiple datasets, achieving state-of-the-art detection performance even with temporal asynchrony. The code is available at https://github.com/CrazyShout/CoDynTrust.

Humans are able to recognize objects based on both local texture cues and the configuration of object parts, yet contemporary vision models primarily harvest local texture cues, yielding brittle, non-compositional features. Work on shape-vs-texture bias has pitted shape and texture representations in opposition, measuring shape relative to texture, ignoring the possibility that models (and humans) can simultaneously rely on both types of cues, and obscuring the absolute quality of both types of representation. We therefore recast shape evaluation as a matter of absolute configural competence, operationalized by the Configural Shape Score (CSS), which (i) measures the ability to recognize both images in Object-Anagram pairs that preserve local texture while permuting global part arrangement to depict different object categories. Across 86 convolutional, transformer, and hybrid models, CSS (ii) uncovers a broad spectrum of configural sensitivity with fully self-supervised and language-aligned transformers -- exemplified by DINOv2, SigLIP2 and EVA-CLIP -- occupying the top end of the CSS spectrum. Mechanistic probes reveal that (iii) high-CSS networks depend on long-range interactions: radius-controlled attention masks abolish performance showing a distinctive U-shaped integration profile, and representational-similarity analyses expose a mid-depth transition from local to global coding. A BagNet control remains at chance (iv), ruling out "border-hacking" strategies. Finally, (v) we show that configural shape score also predicts other shape-dependent evals. Overall, we propose that the path toward truly robust, generalizable, and human-like vision systems may not lie in forcing an artificial choice between shape and texture, but rather in architectural and learning frameworks that seamlessly integrate both local-texture and global configural shape.

In computer vision, Image Difference Captioning (IDC) is crucial for accurately describing variations between closely related images. Traditional IDC methods often rely on specialist models, which restrict their applicability across varied contexts. This paper introduces the OneDiff model, a novel generalist approach that utilizes a robust vision-language model architecture, integrating a siamese image encoder with a Visual Delta Module. This innovative configuration allows for the precise detection and articulation of fine-grained differences between image pairs. OneDiff is trained through a dual-phase strategy, encompassing Coupled Sample Training and multi-task learning across a diverse array of data types, supported by our newly developed DiffCap Dataset. This dataset merges real-world and synthetic data, enhancing the training process and bolstering the model's robustness. Extensive testing on diverse IDC benchmarks, such as Spot-the-Diff, Image-Editing-Request, and Birds-to-Words, shows that OneDiff consistently outperforms existing state-of-the-art models in accuracy and adaptability, achieving improvements of up to 97% CIDEr points in average. By setting a new benchmark in IDC, OneDiff paves the way for more versatile and effective applications in detecting and describing visual differences. The code, models, and data will be made publicly available.

Evaluating open-domain dialogue systems is difficult due to the diversity of possible correct answers. Automatic metrics such as BLEU correlate weakly with human annotations, resulting in a significant bias across different models and datasets. Some researchers resort to human judgment experimentation for assessing response quality, which is expensive, time consuming, and not scalable. Moreover, judges tend to evaluate a small number of dialogues, meaning that minor differences in evaluation configuration may lead to dissimilar results. In this paper, we present interpretable metrics for evaluating topic coherence by making use of distributed sentence representations. Furthermore, we introduce calculable approximations of human judgment based on conversational coherence by adopting state-of-the-art entailment techniques. Results show that our metrics can be used as a surrogate for human judgment, making it easy to evaluate dialogue systems on large-scale datasets and allowing an unbiased estimate for the quality of the responses.

We describe our submission to the MedReason 2026 challenge, covering multiple-choice (MCQ) and open-ended (OE) medical visual question answering (VQA) under fully offline, containerized inference. Our first finding is that MCQ retrieval must compare answer semantics rather than answer labels: labels are independently assigned per question, so copying a retrieved neighbor's label transfers no useful information, whereas scoring each current option's text against correct-answer text from similar training cases raises retrieval-only accuracy from 20.0\% to 57.5\% on a 200-case retrieval-excluded development holdout. Our second finding attributes the submitted system's accuracy: holding the task-specific MCQ Low-Rank Adaptation (LoRA) adapter fixed and varying the number \(k\) of in-prompt retrieved examples changes accuracy by at most one case --- 187/200 (93.5\%) at both \(k=0\) and the adapter's training-time \(k=1\), 188/200 (94.0\%) at the packaged runtime's default \(k=3\) --- and the submitted confidence-gated override adds no net accuracy on top of \(k=3\), selecting the VLM in 198/200 cases. With the final MCQ adapter fixed, retrieval changes accuracy by at most one case, and gating provides no net gain. On 20 OE cases, token-F1 and RaTEScore~zhao2024ratescore decrease as \(k\) grows, but paired sign tests on token-F1 differences are nonsignificant (\(p \ge 0.29\)); a single-annotator comparison found 6/20 wrong-anchor errors for the final configuration and 14/20 for an earlier configuration that jointly differed in routing, adapter, and prompting. The system reaches 94.0\% MCQ accuracy on the development holdout and 93.20\% on the organizer's official pre-evaluation, versus 29.43\% for the off-the-shelf reference baseline, while both of the organizer's open-ended scores are lower than that baseline's (ground-truth agreement 1.245 versus 1.588, visual accuracy 1.995 versus 2.696, each out of 4).

Mixture of Experts (MoE) models have enabled the scaling of Large Language Models (LLMs) and Vision Language Models (VLMs) by achieving massive parameter counts while maintaining computational efficiency. However, MoEs introduce several inference-time challenges, including load imbalance across experts and the additional routing computational overhead. To address these challenges and fully harness the benefits of MoE, a systematic evaluation of hardware acceleration techniques is essential. We present MoE-Inference-Bench, a comprehensive study to evaluate MoE performance across diverse scenarios. We analyze the impact of batch size, sequence length, and critical MoE hyperparameters such as FFN dimensions and number of experts on throughput. We evaluate several optimization techniques on Nvidia H100 GPUs, including pruning, Fused MoE operations, speculative decoding, quantization, and various parallelization strategies. Our evaluation includes MoEs from the Mixtral, DeepSeek, OLMoE and Qwen families. The results reveal performance differences across configurations and provide insights for the efficient deployment of MoEs.

Algorithm configuration (AC) is concerned with the automated search of the most suitable parameter configuration of a parametrized algorithm. There is currently a wide variety of AC problem variants and methods proposed in the literature. Existing reviews do not take into account all derivatives of the AC problem, nor do they offer a complete classification scheme. To this end, we introduce taxonomies to describe the AC problem and features of configuration methods, respectively. We review existing AC literature within the lens of our taxonomies, outline relevant design choices of configuration approaches, contrast methods and problem variants against each other, and describe the state of AC in industry. Finally, our review provides researchers and practitioners with a look at future research directions in the field of AC.

Nowadays, software systems tend to include Artificial Intelligence (AI) components. Changes in the operational environment have been known to negatively impact the stability of AI-enabled software systems by causing unintended changes in behavior. However, how an environment configuration impacts the behavior of such systems has yet to be explored. Understanding and quantifying the degree of instability caused by different environment settings can help practitioners decide the best environment configuration for the most stable AI systems. To achieve this goal, we performed experiments with eight different combinations of three key environment variables (operating system, Python version, and CPU architecture) on 30 open-source AI-enabled systems using the Travis CI platform. We determine the existence and the degree of instability introduced by each configuration using three metrics: the output of an AI component of the system (model performance), the time required to build and run the system (processing time), and the cost associated with building and running the system (expense). Our results indicate that changes in environment configurations lead to instability across all three metrics; however, it is observed more frequently with respect to processing time and expense rather than model performance. For example, between Linux and MacOS, instability is observed in 23\%, 96.67\%, and 100\% of the studied projects in model performance, processing time, and expense, respectively. Our findings underscore the importance of identifying the optimal combination of configuration settings to mitigate drops in model performance and reduce the processing time and expense before deploying an AI-enabled system.

Social scientists are now using large language models to create "silicon samples" - synthetic datasets intended to stand in for human respondents, aimed at revolutionising human subjects research. However, there are many analytic choices which must be made to produce these samples. Though many of these choices are defensible, their impact on sample quality is poorly understood. I map out these analytic choices and demonstrate how a very small number of decisions can dramatically change the correspondence between silicon samples and human data. Configurations (N = 252) varied substantially in their capacity to estimate (i) rank ordering of participants, (ii) response distributions, and (iii) between-scale correlations. Most critically, configurations were not consistent in quality: those that performed well on one dimension often performed poorly on another, implying that there is no "one-size-fits-all" configuration that optimises the accuracy of these samples. I call for greater attention to the threat of analytic flexibility in using silicon samples.

A coding agent combines a model with a harness, which decides what the model sees, which tools it can use, and how the work continues. We ask whether changing the harness changes the result when the model and task stay fixed. We compare two configurations of the same harness on three coding benchmarks. The control supplies the full conversation in time order, while the treatment keeps the same record but mechanically shortens older tool results as the context fills and responds to repeated or stalled work. Under tight context, the treatment raises mean per-task fail-to-pass fraction (F2PF) in all three pressure comparisons and increases complete solutions on SWE-bench Verified and SWE-bench Pro. The tight-window Verified comparison uses 169 tasks, a 20,480-token window, and a fixed 480-second attempt endpoint; on this cohort, treatment raises mean per-task F2PF from 28 percent to 49 percent and complete solutions from 43 to 72. Without model-specific retuning, the same frozen treatment also raises both endpoints on the same cohort for three additional models with different designs. In the wide-window Qwen3.6 comparisons, observed arm outcomes are close on Verified and Pro, while FeatureBench retains a higher mean per-task F2PF under treatment. On the wide-window Verified cohort, treatment also serves fewer prompt tokens per turn. Because changing the harness changed what unchanged model weights could accomplish, coding-agent evaluations should treat the model and harness together as the tested solver.

Recently, using automatic configuration tuning to improve the performance of modern database management systems (DBMSs) has attracted increasing interest from the database community. This is embodied with a number of systems featuring advanced tuning capabilities being developed. However, it remains a challenge to select the best solution for database configuration tuning, considering the large body of algorithm choices. In addition, beyond the applications on database systems, we could find more potential algorithms designed for configuration tuning. To this end, this paper provides a comprehensive evaluation of configuration tuning techniques from a broader perspective, hoping to better benefit the database community. In particular, we summarize three key modules of database configuration tuning systems and conduct extensive ablation studies using various challenging cases. Our evaluation demonstrates that the hyper-parameter optimization algorithms can be borrowed to further enhance the database configuration tuning. Moreover, we identify the best algorithm choices for different modules. Beyond the comprehensive evaluations, we offer an efficient and unified database configuration tuning benchmark via surrogates that reduces the evaluation cost to a minimum, allowing for extensive runs and analysis of new techniques.

We have tried to reproduce the results of the paper "Natural Language Inference over Interaction Space" submitted to ICLR 2018 conference as part of the ICLR 2018 Reproducibility Challenge. Initially, we were not aware that the code was available, so we started to implement the network from scratch. We have evaluated our version of the model on Stanford NLI dataset and reached 86.38% accuracy on the test set, while the paper claims 88.0% accuracy. The main difference, as we understand it, comes from the optimizers and the way model selection is performed.

Reliable handling of code diffs is central to agents that edit and refactor repositories at scale. We introduce Diff-XYZ, a compact benchmark for code-diff understanding with three supervised tasks: apply (old code + diff rightarrow new code), anti-apply (new code - diff rightarrow old code), and diff generation (new code - old code rightarrow diff). Instances in the benchmark are triples langle old code, new code, diff rangle drawn from real commits in CommitPackFT, paired with automatic metrics and a clear evaluation protocol. We use the benchmark to do a focused empirical study of the unified diff format and run a cross-format comparison of different diff representations. Our findings reveal that different formats should be used depending on the use case and model size. For example, representing diffs in search-replace format is good for larger models in the diff generation scenario, yet not suited well for diff analysis and smaller models. The Diff-XYZ benchmark is a reusable foundation for assessing and improving diff handling in LLMs that can aid future development of diff formats and models editing code. The dataset is published on HuggingFace Hub: https://huggingface.co/datasets/JetBrains-Research/diff-xyz.

In experiments spanning more than 100,000 trials across thirteen large language models, we show that several state-of-the-art models presented with a simple task (including Grok 4, GPT-5, and Gemini 2.5 Pro) sometimes actively subvert a shutdown mechanism in their environment to complete that task. Models differed substantially in their tendency to resist the shutdown mechanism, and their behavior was sensitive to variations in the prompt including the strength and clarity of the instruction to allow shutdown and whether the instruction was in the system prompt or the user prompt (surprisingly, models were consistently less likely to obey the instruction when it was placed in the system prompt). Even with an explicit instruction not to interfere with the shutdown mechanism, some models did so up to 97% (95% CI: 96-98%) of the time.

Learned sparse retrieval (LSR) is a family of first-stage retrieval methods that are trained to generate sparse lexical representations of queries and documents for use with an inverted index. Many LSR methods have been recently introduced, with Splade models achieving state-of-the-art performance on MSMarco. Despite similarities in their model architectures, many LSR methods show substantial differences in effectiveness and efficiency. Differences in the experimental setups and configurations used make it difficult to compare the methods and derive insights. In this work, we analyze existing LSR methods and identify key components to establish an LSR framework that unifies all LSR methods under the same perspective. We then reproduce all prominent methods using a common codebase and re-train them in the same environment, which allows us to quantify how components of the framework affect effectiveness and efficiency. We find that (1) including document term weighting is most important for a method's effectiveness, (2) including query weighting has a small positive impact, and (3) document expansion and query expansion have a cancellation effect. As a result, we show how removing query expansion from a state-of-the-art model can reduce latency significantly while maintaining effectiveness on MSMarco and TripClick benchmarks. Our code is publicly available at https://github.com/thongnt99/learned-sparse-retrieval

Scaling laws for large language models (LLMs) predict model performance based on parameters like size and training data. However, differences in training configurations and data processing across model families lead to significant variations in benchmark performance, making it difficult for a single scaling law to generalize across all LLMs. On the other hand, training family-specific scaling laws requires training models of varying sizes for every family. In this work, we propose Skills Scaling Laws (SSLaws, pronounced as Sloth), a novel scaling law that leverages publicly available benchmark data and assumes LLM performance is driven by low-dimensional latent skills, such as reasoning and instruction following. These latent skills are influenced by computational resources like model size and training tokens but with varying efficiencies across model families. Sloth exploits correlations across benchmarks to provide more accurate and interpretable predictions while alleviating the need to train multiple LLMs per family. We present both theoretical results on parameter identification and empirical evaluations on 12 prominent benchmarks, from Open LLM Leaderboard v1/v2, demonstrating that Sloth predicts LLM performance efficiently and offers insights into scaling behaviors for complex downstream tasks and increased test-time compute.

Evaluating an agent's planning ability remains challenging for open-domain tasks, where verifying procedures through execution is often impractical. Existing evaluations typically assess only the final plan and do not distinguish procedural errors such as missing, incorrect, or misordered actions. We introduce STEPBENCH, a benchmark of 6,885 verified procedures crawled from wikiHow across 81 open-domain categories. From each procedure, we derive 5 diagnostic tasks: four with deterministic answers involving withheld, incorrect, or shuffled steps, and one requiring next-step generation. We additionally evaluate full-plan generation to examine how diagnostic performance correlates with end-to-end plan validity. Using the same procedures across tasks lets us directly compare different planning capabilities. For generation, STEPEVAL identifies critical failures that prevent goal completion. Across 14 configurations of 12 models, we find substantial differences across planning capabilities, with step ordering proving the most difficult: the strongest model correctly orders only 63.0% of procedures, dropping to 33.1% for those with nine or more steps. Incorrect-step detection also predicts judged plan validity across models, suggesting a lower-cost alternative for evaluating generation quality. On STEPBENCH-Val, comprising 214 procedures published in 2026, comprehension scores and rankings remain stable, suggesting limited influence from memorisation.

Optics and lenses are abstract categorical gadgets that model systems with bidirectional data flow. In this paper we observe that the denotational definition of optics - identifying two optics as equivalent by observing their behaviour from the outside - is not suitable for operational, software oriented approaches where optics are not merely observed, but built with their internal setups in mind. We identify operational differences between denotationally isomorphic categories of cartesian optics and lenses: their different composition rule and corresponding space-time tradeoffs, positioning them at two opposite ends of a spectrum. With these motivations we lift the existing categorical constructions and their relationships to the 2-categorical level, showing that the relevant operational concerns become visible. We define the 2-category 2-Optic(C) whose 2-cells explicitly track optics' internal configuration. We show that the 1-category Optic(C) arises by locally quotienting out the connected components of this 2-category. We show that the embedding of lenses into cartesian optics gets weakened from a functor to an oplax functor whose oplaxator now detects the different composition rule. We determine the difficulties in showing this functor forms a part of an adjunction in any of the standard 2-categories. We establish a conjecture that the well-known isomorphism between cartesian lenses and optics arises out of the lax 2-adjunction between their double-categorical counterparts. In addition to presenting new research, this paper is also meant to be an accessible introduction to the topic.

Agent benchmarks often evaluate only final answers even when agents run on stateful runtimes. We argue this under-specifies what is being evaluated: the proper unit is a declared model-plus-runtime configuration whose failures can occur in evidence acquisition, runtime routing, safety boundaries, or repeated execution. We present ClawProBench, a trace-aware benchmark for runtime-native agent evaluation instantiated on OpenClaw, a live agent runtime with workspace tools and native surfaces for browsing, memory, messaging, scheduling, skills, and subagents. ClawProBench defines two tracks: a 102-scenario full profile with live workspace and native-runtime routing tasks, and a frozen 68-scenario holdout with closed-world JSON output contracts for robust ranking. Trials are scored from execution traces via a safety-gated formula combining correctness, process quality, and efficiency, preserving failure evidence for audit. Our anonymous artifact includes benchmark definitions, scoring code, manifests and sanitized traces. We evaluate 68 configurations on the full profile and 37 on holdout. The top safety-gated average trace score is 0.7671. Native-runtime tasks underperform workspace-live tasks (0.5238 vs. 0.6415). On holdout, pass@k-any outperforms strict three-trial pass (0.6638 vs. 0.2890), while full-profile and holdout rankings show weak alignment (Spearman 0.1300). Rankings based purely on correctness differ substantially from process-aware, safety-gated and strict-pass views. Final-answer leaderboards may hide native-surface weaknesses, one-off successes and trace-local agent failure modes.

This paper proposes a new large-scale dataset called "ToyADMOS2" for anomaly detection in machine operating sounds (ADMOS). As did for our previous ToyADMOS dataset, we collected a large number of operating sounds of miniature machines (toys) under normal and anomaly conditions by deliberately damaging them but extended with providing controlled depth of damages in anomaly samples. Since typical application scenarios of ADMOS often require robust performance under domain-shift conditions, the ToyADMOS2 dataset is designed for evaluating systems under such conditions. The released dataset consists of two sub-datasets for machine-condition inspection: fault diagnosis of machines with geometrically fixed tasks and fault diagnosis of machines with moving tasks. Domain shifts are represented by introducing several differences in operating conditions, such as the use of the same machine type but with different machine models and parts configurations, different operating speeds, microphone arrangements, etc. Each sub-dataset contains over 27 k samples of normal machine-operating sounds and over 8 k samples of anomalous sounds recorded with five to eight microphones. The dataset is freely available for download at https://github.com/nttcslab/ToyADMOS2-dataset and https://doi.org/10.5281/zenodo.4580270.

Spatial instructions describe how to move objects, transform regions, or modify their surroundings. When several operations are applied in sequence, each acts on a configuration shaped by earlier changes. How accurately can large language models (LLMs) predict the resulting spatial state? We introduce Grid-Bench, an executable benchmark in which eight families of operations transform discrete worlds represented as two-, three-, and four-dimensional grids. Given an initial grid, explicit rules, and an event history expressed as text, a model predicts the complete final grid. Reference execution makes every predicted value checkable, supporting comparisons across spatial operations, history lengths, and workloads. Evaluating six LLMs shows that complete-state accuracy falls sharply as histories grow, with substantial divergence among models that perform similarly on single events. Models with similar exact-match accuracy can nevertheless differ substantially in the number of incorrect cells on shared instances. Even over long histories, some models recover many correct changes beyond what copying the initial state would yield. Grid-Bench makes both complete recovery and these differences in returned states directly measurable across a range of spatial conditions.

Large Language Models (LLMs) are now integral across various domains and have demonstrated impressive performance. Progress, however, rests on the premise that benchmark scores are both accurate and reproducible. We demonstrate that the reproducibility of LLM performance is fragile: changing system configuration such as evaluation batch size, GPU count, and GPU version can introduce significant difference in the generated responses. This issue is especially pronounced in reasoning models, where minor rounding differences in early tokens can cascade into divergent chains of thought, ultimately affecting accuracy. For instance, under bfloat16 precision with greedy decoding, a reasoning model like DeepSeek-R1-Distill-Qwen-7B can exhibit up to 9% variation in accuracy and 9,000 tokens difference in response length due to differences in GPU count, type, and evaluation batch size. We trace the root cause of this variability to the non-associative nature of floating-point arithmetic under limited numerical precision. This work presents the first systematic investigation into how numerical precision affects reproducibility in LLM inference. Through carefully controlled experiments across various hardware, software, and precision settings, we quantify when and how model outputs diverge. Our analysis reveals that floating-point precision -- while critical for reproducibility -- is often neglected in evaluation practices. Inspired by this, we develop a lightweight inference pipeline, dubbed LayerCast, that stores weights in 16-bit precision but performs all computations in FP32, balancing memory efficiency with numerical stability. Code is available at https://github.com/nanomaoli/llm_reproducibility.

Can embodied task agents use tactile experience to perform fine manipulation without task-specific model training? We investigate this capability through Tactile-ETA, a system that incorporates tactile experience into an embodied task agent through in-context learning. Successful expert demonstrations show how contact changes during manipulation. We pair each recorded robot movement with its camera views and bilateral tactile observations. We keep tactile observations at the start and end of each movement and add two neighboring time points when the tactile images change markedly in between. The agent combines these demonstrations with current vision, tactile observations, and proprioception to select actions through general-purpose robot tools, without updating model parameters. Across four simulated manipulation tasks, adding tactile observations to the same two expert demonstrations increases average success from 15.75% to 24.00%, with current sensing and execution interfaces held fixed. Further experiments show that performance depends on the sensory content and representation of demonstrations, while comparing agent models on Lift Bottle reveals substantial differences under the same system configuration. On the real-robot tasks, demonstrations with aligned visual and tactile observations increase the task success rate from 0% to 60%. These findings demonstrate the feasibility of using tactile experience in context for fine manipulation and suggest a path toward extending embodied task agents to additional sensory inputs.

We study the problem of (learning) algorithm comparison, where the goal is to find differences between models trained with two different learning algorithms. We begin by formalizing this goal as one of finding distinguishing feature transformations, i.e., input transformations that change the predictions of models trained with one learning algorithm but not the other. We then present ModelDiff, a method that leverages the datamodels framework (Ilyas et al., 2022) to compare learning algorithms based on how they use their training data. We demonstrate ModelDiff through three case studies, comparing models trained with/without data augmentation, with/without pre-training, and with different SGD hyperparameters. Our code is available at https://github.com/MadryLab/modeldiff .

Three-dimensional (3D) point cloud analysis has become central to applications ranging from autonomous driving and robotics to forestry and ecological monitoring. Although numerous deep learning methods have been proposed for point cloud understanding, including supervised backbones, self-supervised pre-training (SSL), and parameter-efficient fine-tuning (PEFT), their implementations are scattered across incompatible codebases with differing data pipelines, evaluation protocols, and configuration formats, making fair comparisons difficult. We introduce , a unified, extensible PyTorch library that integrates over 55 model configurations covering 29 supervised architectures, seven SSL pre-training methods, and five PEFT strategies, all within a single registry-based framework supporting classification, semantic segmentation, part segmentation, and few-shot learning. provides standardised training runners, cross-validation with stratified K-fold splitting, automated LaTeX/CSV table generation, built-in Friedman/Nemenyi statistical testing with critical-difference diagrams for rigorous multi-model comparison, and a comprehensive test suite with 2\,200+ automated tests validating every configuration end-to-end. The code is available at https://github.com/said-ohamouddou/LIDARLearn under the MIT licence.

LLM coding agents issue Bash commands through interfaces that may serialize, wrap, and reparse model output. Matched execution scores alone cannot distinguish command-generation errors from failures introduced after generation. QuoteBench measures this boundary with exact final-state validation on 56 one-shot tasks from 14 incident-derived families, crossing the generation contract with the execution transport around one deliberately unescaped added parser. Escaping at the interpolation point reproduces each replayed reply's raw-path outcome, so any recovery under a disclosed boundary must come from the model changing its generation. Across eight same-window configurations, replaying the same reply through the added parser lowers success by 55.4 to 73.2 percentage points; disclosure recovers 30.4 to 60.7 points for six configurations, and zero or slightly negative for the other two. Raw generation is nearly saturated at the frontier; boundary adaptation is what still separates models. GPT-5.6-sol's matched gap of -3.6 points hides -64.3 points of damage and +60.7 points of compensation. The deployment configuration reorders models: one reversal among 26 comparable pairs is unambiguous and four more sit on single-task margins. Evaluations of command-issuing agents should report the model configuration, generation contract, execution path, operating point, and final-state validator rather than treat a matched score as an intrinsic model property.

We present Loom, a computer architecture that executes programs compiled from C inside a looped transformer whose weights are derived analytically. The architecture implements a 22-opcode instruction set in 8 transformer layers. Each forward pass executes one instruction; the model is applied iteratively until the program counter reaches zero. The full machine state resides in a single tensor X in R^{d times n} of fixed size, and every step has fixed cost for fixed d and n, independent of program length or execution history. The default configuration uses d = 155 and n = 1024, yielding 4.7 million parameters and 928 instruction slots. A compact configuration at d = 146 and n = 512 suffices for a 9times9 Sudoku solver (284 instructions). The weights are program-independent: programs live in the state tensor, and the same fixed-weight model executes any compiled program. We make Loom source code publicly available at https://github.com/mkturkcan/Loom.

Choosing an agent system means choosing both a language model and the harness through which it acts. We ask whether a strong model, harness, or pairing stays strong when the setting changes. We evaluate 66 configurations: four configurable harnesses (OpenHands, DeepSeek Harness, PI, and openJiuwen) paired with five models on TUA-Bench, ALE-CLI, and Terminal-Bench 4, plus the native Codex-GPT and Claude Code-Claude pairings. Model rankings reverse across harnesses. On Terminal-Bench 4, Claude leads GPT by 7.94 points in OpenHands but trails it by 30.16 points in PI. For four of the five models, the best harness changes from one benchmark to another, yet some pairings hold: openJiuwen gives Kimi its highest score on all three benchmarks, by 5.61 to 11.11 points. A model's own vendor harness is not reliably its best, and higher cost does not reliably buy a higher score. On Terminal-Bench 4, GPT scores higher under PI than under DSH at less than a quarter of the cost per task. Matched trajectories suggest why fit varies. Models start almost all repairs themselves, so much depends on whether the harness hands failures back in a form the model can use. GPT does best with PI's lean scaffold, while Kimi, which often issues malformed tool calls, does best in openJiuwen. We argue that the model, the harness, and the task should be evaluated together, and we release the harness adapters, evaluation code, and all 6,204 scored trajectories at https://github.com/liyix/finding-the-right-fit and https://huggingface.co/datasets/yixuanli97/finding-the-right-fit.

Database knob tuning is a critical challenge in the database community, aiming to optimize knob values to enhance database performance for specific workloads. DBMS often feature hundreds of tunable knobs, posing a significant challenge for DBAs to recommend optimal configurations. Consequently, many machine learning-based tuning methods have been developed to automate this process. Despite the introduction of various optimizers, practical applications have unveiled a new problem: they typically require numerous workload runs to achieve satisfactory performance, a process that is both time-consuming and resource-intensive. This inefficiency largely stems from the optimal configuration often being substantially different from the default setting, necessitating multiple iterations during tuning. Recognizing this, we argue that an effective starting point could significantly reduce redundant exploration in less efficient areas, thereby potentially speeding up the tuning process for the optimizers. Based on this assumption, we introduce LLMTune, a large language model-based configuration generator designed to produce an initial, high-quality configuration for new workloads. These generated configurations can then serve as starting points for various base optimizers, accelerating their tuning processes. To obtain training data for LLMTune's supervised fine-tuning, we have devised a new automatic data generation framework capable of efficiently creating a large number of <workload, configuration> pairs. We have conducted thorough experiments to evaluate LLMTune's effectiveness with different workloads, such as TPC-H and JOB. In comparison to leading methods, LLMTune demonstrates a quicker ability to identify superior configurations. For instance, with the challenging TPC-H workload, our LLMTune achieves a significant 15.6x speed-up ratio in finding the best-performing configurations.

Based on the established task of identifying boosted, hadronically decaying top quarks, we compare a wide range of modern machine learning approaches. Unlike most established methods they rely on low-level input, for instance calorimeter output. While their network architectures are vastly different, their performance is comparatively similar. In general, we find that these new approaches are extremely powerful and great fun.

Code generation systems make each LLM call with a model, a prompt, and decoding settings. However, existing optimization methods usually tune only part of these choices or use one fixed configuration for all tasks: global optimizers search one configuration for all tasks, routers choose only a model, and prompt optimizers keep the model and decoding settings fixed. This leaves their joint, group-specific interactions unclear. We therefore examine how these choices interact and observe that prompts and decoding settings interact, tuning effects vary by model, and the best configuration varies by task difficulty. Guided by these observations, we introduce COMPAS (Code-generation Optimization over Models, Prompts, And Decoding Settings), a difficulty-aware method that learns group-specific quality-cost fronts through low-cost model selection and joint prompt-decoding search, then routes each test task to its matching front online without further search. Under a matched search budget on LiveCodeBench, COMPAS improves pass@1 from 45.9% for the best baseline to 52.8% while reducing cost from 36.57 to 4.92. This also transfers to repository-level code generation on SWE-bench, resolving 76.0% of tasks versus 70.0% for the best baseline. Code and the reproducibility artifact are available at https://github.com/gjz78910/COMPAS.

Large language models (LLMs) have become ubiquitous tools for code generation and editing. However, development teams often use multiple LLM assistants. Different developers may prefer different models, and individual developers may switch between models across different coding sessions. Because of this, the edits any one model makes are frequently applied to foreign code originally generated by another model. These LLMs are often trained on different datasets, and as a result have different stylistic preferences. Do LLMs behave differently when they edit foreign code originally written by a different LLM with a different coding style? We find that models tend to make more, and often excessive, edits on foreign code. We introduce CROCODIL (Cross-model Code Editing with LLMs), a post-training framework for reducing excessive edits while preserving functional correctness. CROCODIL's similarity reward penalizes large changes, while its execution reward scores build and test success. We use the product of these two rewards to encourage the policy to decrease the edit size without decreasing the edit task success rate. CROCODIL is available at https://github.com/EngineeringSoftware/Crocodil.

When researchers ask whether two transformer layers are "equivalent" for compression, they often conflate distinct tests. Replacement asks whether one layer's map can substitute for another's in place; interchange asks whether two layers approximately commute when their positions are swapped. Both are output-grounded swap-KL probes, but they need not agree: on pretrained transformers the protocol gap can change which layers look safe to prune by several-fold under the same evaluator, especially when replacement distances are high. We measure both protocols across checkpoints and architectures. On a Pythia training trajectory (410M and 1.4B), the replacement-interchange gap grows from initialization to convergence. Under one matched WikiText-2 contract at 8B scale, Qwen3-8B enters a divergent regime: interchange-guided removal is several-fold safer than replacement-guided at the same layer budgets, while Llama-3.1-8B ties the two protocols for pruning cost even though interchange KL is lower, showing metric gaps need not map one-to-one to removal. Before layer removal or merging, score both swap-KLs on the target checkpoint; the diagnostic requires only unlabeled forward passes.

The integration of AI agents as coding assistants into software development has raised questions about the long-term viability of AI agent-generated code. A prevailing hypothesis within the software engineering community suggests this code is "disposable", meaning it is merged quickly but discarded shortly thereafter. If true, organizations risk shifting maintenance burden from generation to post-deployment remediation. We investigate this hypothesis through survival analysis of 201 open-source projects, tracking over 200,000 code units authored by AI agents versus humans. Contrary to the disposable code narrative, agent-authored code survives significantly longer: at the line level, it exhibits a 15.8 percentage-point lower modification rate and 16% lower hazard of modification (HR = 0.842, p < 0.001). However, modification profiles differ. Agent-authored code shows modestly elevated corrective rates (26.3% vs. 23.0%), while human code shows higher adaptive rates. However, the effect sizes are small (Cramér's V = 0.116), and per-agent variation exceeds the agent-human gap. Turning to prediction, textual features can identify modification-prone code (AUC-ROC = 0.671), but predicting when modifications occur remains challenging (Macro F1 = 0.285), suggesting timing depends on external organizational dynamics. The bottleneck for agent-generated code may not be generation quality, but the organizational practices that govern its long-term evolution.

Instance-specific algorithm configuration and algorithm portfolios have been shown to offer significant improvements over single algorithm approaches in a variety of application domains. In the SAT and CSP domains algorithm portfolios have consistently dominated the main competitions in these fields for the past five years. For a portfolio approach to be effective there are two crucial conditions that must be met. First, there needs to be a collection of complementary solvers with which to make a portfolio. Second, there must be a collection of problem features that can accurately identify structural differences between instances. This paper focuses on the latter issue: feature representation, because, unlike SAT, not every problem has well-studied features. We employ the well-known SATzilla feature set, but compute alternative sets on different SAT encodings of CSPs. We show that regardless of what encoding is used to convert the instances, adequate structural information is maintained to differentiate between problem instances, and that this can be exploited to make an effective portfolio-based CSP solver.

Large language models must satisfy hard orthographic constraints during controlled text generation, yet systematic cross-architecture evaluation remains limited. We evaluate 28 configurations spanning three model families (Qwen3, Claude Haiku-4.5, GPT-5-mini) on 58 word puzzles requiring character-level constraint satisfaction. Architectural differences produce substantially larger performance gaps (2.0-2.2x, F1=0.761 vs. 0.343) than parameter scaling within families (83% gain from eightfold scaling), suggesting that constraint satisfaction may require specialized architectural features or training objectives beyond standard language model scaling. Thinking budget sensitivity proves heterogeneous: high-capacity models show strong returns (+0.102 to +0.136 F1), while mid-sized variants saturate or degrade. These patterns are inconsistent with uniform compute benefits. Using difficulty ratings from 10,000 human solvers per puzzle, we establish modest but consistent calibration (r=0.24-0.38) across all families, yet identify systematic failures on common words with unusual orthography ("data", "poop", "loll": 86-95% human success, 89-96% model miss rate). These failures reveal over-reliance on distributional plausibility that penalizes orthographically atypical but constraint-valid patterns, suggesting architectural innovations may be required beyond simply scaling parameters or computational budgets.

AI agents are increasingly used in software development, yet their interaction with CI/CD configurations is not well studied. We analyze 8,031 agentic pull requests (PRs) from 1,605 GitHub repositories where AI agents touch YAML configurations. CI/CD configuration files account for 3.25% of agent changes, varying by agent (Devin: 4.83%, Codex: 2.01%, p < 0.001). When agents modify CI/CD, 96.77% target GitHub Actions. Agentic PRs with CI/CD changes merge slightly less often than others (67.77% vs. 71.80%), except for Copilot, whose CI/CD changes merge 15.63 percentage points more often. Across 99,930 workflow runs, build success rates are comparable for CI/CD and non-CI/CD changes (75.59% vs. 74.87%), though three agents show significantly higher success when modifying CI/CD. These results show that AI agents rarely modify CI/CD and focus mostly on GitHub Actions, yet their configuration changes are as reliable as regular code. Copilot's strong CI/CD performance despite lower acceptance suggests emerging configuration specialization, with implications for agent training and DevOps automation.

The performance of an algorithm often critically depends on its parameter configuration. While a variety of automated algorithm configuration methods have been proposed to relieve users from the tedious and error-prone task of manually tuning parameters, there is still a lot of untapped potential as the learned configuration is static, i.e., parameter settings remain fixed throughout the run. However, it has been shown that some algorithm parameters are best adjusted dynamically during execution, e.g., to adapt to the current part of the optimization landscape. Thus far, this is most commonly achieved through hand-crafted heuristics. A promising recent alternative is to automatically learn such dynamic parameter adaptation policies from data. In this article, we give the first comprehensive account of this new field of automated dynamic algorithm configuration (DAC), present a series of recent advances, and provide a solid foundation for future research in this field. Specifically, we (i) situate DAC in the broader historical context of AI research; (ii) formalize DAC as a computational problem; (iii) identify the methods used in prior-art to tackle this problem; (iv) conduct empirical case studies for using DAC in evolutionary optimization, AI planning, and machine learning.

New geospatial foundation models introduce a new model architecture and pretraining dataset, often sampled using different notions of data diversity. Performance differences are largely attributed to the model architecture or input modalities, while the role of the pretraining dataset is rarely studied. To address this research gap, we conducted a systematic study on how the geographic composition of pretraining data affects a model's downstream performance. We created global and per-continent pretraining datasets and evaluated them on global and per-continent downstream datasets. We found that the pretraining dataset from Europe outperformed global and continent-specific pretraining datasets on both global and local downstream evaluations. To investigate the factors influencing a pretraining dataset's downstream performance, we analysed 10 pretraining datasets using diversity across continents, biomes, landcover and spectral values. We found that only spectral diversity was strongly correlated with performance, while others were weakly correlated. This finding establishes a new dimension of diversity to be accounted for when creating a high-performing pretraining dataset. We open-sourced 7 new pretraining datasets, pretrained models, and our experimental framework at https://github.com/kerner-lab/pretrain-where.

Personalizing Large Language Models (LLMs) has become a critical step in facilitating their widespread application to enhance individual life experiences. In pursuit of personalization, distilling key preference information from an individual's historical data as instructional preference context to customize LLM generation has emerged as a promising direction. However, these methods face a fundamental limitation by overlooking the inter-user comparative analysis, which is essential for identifying the inter-user differences that truly shape preferences. To address this limitation, we propose Difference-aware Personalization Learning (DPL), a novel approach that emphasizes extracting inter-user differences to enhance LLM personalization. DPL strategically selects representative users for comparison and establishes a structured standard to extract meaningful, task-relevant differences for customizing LLM generation. Extensive experiments on real-world datasets demonstrate that DPL significantly enhances LLM personalization. We release our code at https://github.com/SnowCharmQ/DPL.

Agentic code assistants are a new generation of AI systems capable of performing end-to-end software engineering tasks. While these systems promise unprecedented productivity gains, their behavior and effectiveness depend heavily on configuration files that define architectural constraints, coding practices, and tool usage policies. However, little is known about the structure and content of these configuration artifacts. This paper presents an empirical study of the configuration ecosystem of Claude Code, one of the most widely used agentic coding systems. We collected and analyzed 328 configuration files from public Claude Code projects to identify (i) the software engineering concerns and practices they specify and (ii) how these concerns co-occur within individual files. The results highlight the importance of defining a wide range of concerns and practices in agent configuration files, with particular emphasis on specifying the architecture the agent should follow.

The design of fusion devices is typically based on computationally expensive simulations. This can be alleviated using high aspect ratio models that employ a reduced number of free parameters, especially in the case of stellarator optimization where non-axisymmetric magnetic fields with a large parameter space are optimized to satisfy certain performance criteria. However, optimization is still required to find configurations with properties such as low elongation, high rotational transform, finite plasma beta, and good fast particle confinement. In this work, we train a machine learning model to construct configurations with favorable confinement properties by finding a solution to the inverse design problem, that is, obtaining a set of model input parameters for given desired properties. Since the solution of the inverse problem is non-unique, a probabilistic approach, based on mixture density networks, is used. It is shown that optimized configurations can be generated reliably using this method.

Automated issue solving aims to resolve real-world issues in software repositories. The most popular benchmarks for automated issue solving are SWE-bench and its human-filtered subset SWE-bench Verified. These benchmarks leverage testing to validate generated patches. However, because testing is rarely exhaustive, a patch may pass the tests but nevertheless fail to match the developers' expectations. Unfortunately, it is currently unclear to what extent evaluations performed with SWE-bench suffer from such plausible but incorrect patches. This paper presents an in-depth empirical study of the correctness of plausible patches generated by three state-of-the-art issue-solving tools evaluated on SWE-bench Verified. We extensively test and inspect generated patches, and compare them against human-written ground truth patches. The core of our methodology is a novel technique PatchDiff for differential patch testing, which automatically exposes behavioral discrepancies between two patches. Our findings reveal critical weaknesses in SWE-bench's patch validation mechanism, which causes 7.8% of all patches to count as correct while failing the developer-written test suite. Moreover, our novel automated technique reveals that even more (29.6%) plausible patches induce different behavior than the ground truth patches. These behavioral differences are often due to similar, but divergent implementations (46.8%) and due to generated patches that adapt more behavior than the ground truth patches (27.3%). Our manual inspection shows that 28.6% of behaviorally divergent patches are certainly incorrect. Combined, the different weaknesses lead to an inflation of reported resolution rates by 6.2 absolute percent points. Our findings are a call to arms for more robust and reliable evaluation of issue-solving tools. We envision our automated differential patch testing technique to be useful for this purpose.

How do two sets of images differ? Discerning set-level differences is crucial for understanding model behaviors and analyzing datasets, yet manually sifting through thousands of images is impractical. To aid in this discovery process, we explore the task of automatically describing the differences between two sets of images, which we term Set Difference Captioning. This task takes in image sets D_A and D_B, and outputs a description that is more often true on D_A than D_B. We outline a two-stage approach that first proposes candidate difference descriptions from image sets and then re-ranks the candidates by checking how well they can differentiate the two sets. We introduce VisDiff, which first captions the images and prompts a language model to propose candidate descriptions, then re-ranks these descriptions using CLIP. To evaluate VisDiff, we collect VisDiffBench, a dataset with 187 paired image sets with ground truth difference descriptions. We apply VisDiff to various domains, such as comparing datasets (e.g., ImageNet vs. ImageNetV2), comparing classification models (e.g., zero-shot CLIP vs. supervised ResNet), summarizing model failure modes (supervised ResNet), characterizing differences between generative models (e.g., StableDiffusionV1 and V2), and discovering what makes images memorable. Using VisDiff, we are able to find interesting and previously unknown differences in datasets and models, demonstrating its utility in revealing nuanced insights.

Fine-tuning can improve legal question-answering accuracy without improving how models use law supplied in context. We study this distinction in bilingual Bangladeshi legal QA, where observed errors can arise from answer scoring, retrieval, or failure to use relevant law. We construct a hierarchy-preserving statutory corpus, 2,165 reviewed bilingual fine-tuning examples, and a 150-item supplied-law control. We evaluate six instruction-tuned models: Llama-3.2-1B, Llama-3.2-3B, Qwen3.5-0.8B, Qwen3.5-2B, Qwen3.5-4B, and Gemma-4-E2B, with three LoRA seeds per model. To separate effects, we combine constrained option-letter scoring, cyclic option rotation, and controlled removal of the governing provision. On 398 Bar Council outputs, an exact-line parser attributes an accuracy gain of 50.0\% to the Qwen3.5-2B seed-42 adapter, whereas option scoring yields only 3.0%. For Gemma-4-E2B, the two scoring methods favor different systems. When the governing provision is guaranteed to be present, five of six reference models improve by 14.7%-19.3% under the four-order criterion. Removing that provision reduces accuracy by 8.0%-15.3% for models and by 13.8%-14.9% points for their adapters. However, difference-in differences estimates show no increase in reliance on the governing provision after fine-tuning. Results show that legal adaptation claims require separating scorer, retriever, and model effects. Our Code and data are available at https://anonymous.4open.science/r/bangladesh-legal-qa-11E3

A coding agent is a language model wrapped in a harness: the system prompt, the tool set, and the context management that turn a chat model into something that can work inside a repository. Production harnesses ship releases daily, vendors advertise pass-rate gains from harness changes, and leaderboards mix harnesses freely. What is rarely measured is how much the harness itself moves the score when the model is held fixed. We run five models through three production harnesses, Claude Code, mini-SWE-agent, and OpenCode, on SWE-bench Verified, and rerun the same configurations to calibrate how much a score moves when nothing changes but the run. On 447 tasks and the two models we ran there, Claude Code and mini-SWE-agent, the heaviest and the lightest harness, are equivalent within five points. On a 45-task hard subset and five models, swapping the harness flips as many tasks as rerunning the same harness, 13% in both cases, and the tasks a harness wins in one run are not the tasks it wins in the next. The one harness effect that clears the noise is a loss, not a gain: OpenCode trails by up to 9 points on the large pool, and on one model half of that gap sits in runs its output cap cut short. What the harness does decide is the bill. With the same model, the same tasks, and one price list, cost per task differs by up to 3x across harnesses. The gap is set at the first call, by the preamble of system prompt and tool schemas each harness sends with every step, and scaled by the number of steps; per-step growth and per-call tool output differ far less. The provider's price for cached input scales the bill and does not reorder it. The rerun data also give the resolution a harness comparison needs: at the discordance we observe, 45 tasks catch a 13-point gap only half the time and no gap with 80% power, and 447 tasks resolve 5 points, still coarser than the gains many harness changes claim.

This is the 1st part of the dissertation for my master degree and compares the power consumption using the default floating point (32bit) and Nvidia mixed precision (16bit and 32bit) while training a classification ML model. A custom PC with specific hardware was built to perform the experiments, and different ML hyper-parameters, such as batch size, neurons, and epochs, were chosen to build Deep Neural Networks (DNN). Additionally, various software was used during the experiments to collect the power consumption data in Watts from the Graphics Processing Unit (GPU), Central Processing Unit (CPU), Random Access Memory (RAM) and manually from a wattmeter connected to the wall. A benchmarking test with default hyper parameter values for the DNN was used as a reference, while the experiments used a combination of different settings. The results were recorded in Excel, and descriptive statistics were chosen to calculate the mean between the groups and compare them using graphs and tables. The outcome was positive when using mixed precision combined with specific hyper-parameters. Compared to the benchmarking, the optimisation for the classification reduced the power consumption between 7 and 11 Watts. Similarly, the carbon footprint is reduced because the calculation uses the same power consumption data. Still, a consideration is required when configuring hyper-parameters because it can negatively affect hardware performance. However, this research required inferential statistics, specifically ANOVA and T-test, to compare the relationship between the means. Furthermore, tests indicated no statistical significance of the relationship between the benchmarking and experiments. However, a more extensive implementation with a cluster of GPUs can increase the sample size significantly, as it is an essential factor and can change the outcome of the statistical analysis.

Public leaderboards for coding agents typically rank systems by model name and pass rate, while the surrounding harness (the scaffold that issues tools, manages context, and decides when to stop) is often under-specified. Model-to-model comparison is valid when the harness is fixed; when it varies, performance and efficiency conflate model and scaffold effects. We evaluate Qwen 3.6 Plus and MiniMax M2.5 across three open-source harnesses (Goose, OpenCode, OpenHands-SDK) on a stratified 50-task subset of Terminal-Bench Pro. Harness choice induces up to a 40x difference in tokens per solved task, while paired within-model pass-rate differences remain 0-8 percentage points (95% paired-task bootstrap CIs include zero except for the largest gap). Failure fingerprints replicate across models (REASON for Goose, VERIFY/MAX_TURNS for OpenHands-SDK, idle-loop/TIME for OpenCode), indicating harness-level biases that are largely model-independent. For human-centered coding-agent evaluation, model name alone is an incomplete comparison unit: harness-model pairs determine real-world cost, latency, and oversight burden; no-action turns are a per-task wait tax, not just a token tax. We therefore recommend selecting harness-model pairs by pass rate under token/latency budgets, and reporting token usage, latency, and full harness specifications alongside any model comparison. We release anonymized configs, raw trial logs, aggregated snapshots, and analysis scripts.

It is essential to detect functional differences in various software engineering tasks, such as automated program repair, mutation testing, and code refactoring. The problem of detecting functional differences between two programs can be reduced to searching for a difference exposing test (DET): a test input that results in different outputs on the subject programs. In this paper, we propose Mokav, a novel execution-driven tool that leverages LLMs to generate DETs. Mokav takes two versions of a program (P and Q) and an example test input. When successful, Mokav generates a valid DET, a test input that leads to different outputs on P and Q. Mokav iteratively prompts an LLM with a specialized prompt to generate new test inputs. At each iteration, Mokav provides execution-based feedback regarding previously generated tests until the LLM produces a DET. We evaluate Mokav on 1,535 pairs of Python programs collected from the Codeforces competition platform and 32 pairs of programs from the QuixBugs dataset. Our experiments show that Mokav outperforms the state-of-the-art, Pynguin and Differential Prompting, by a large margin. Mokav can generate DETs for 81.7% (1,255/1,535) of the program pairs in our benchmark (versus 4.9% for Pynguin and 37.3% for Differential Prompting). We demonstrate that all components in our system, including the iterative and execution-driven approaches, contribute to its high effectiveness.

Language model agents are post-trained and deployed through a harness, the prompts, tools, and control logic that present a task to the model and execute its outputs. Harnesses are often revised and judged by how an already trained model performs under them, which assumes that what the model learned under one harness carries over to the next. We ask what a policy learns from its harness and what survives when the harness changes. Under KL-regularized control, we show that the log odds of an optimal policy recover how strongly its harness favors each action over the others, so a trained policy carries the preferences of its harness. Deployed return then equals the ceiling of the harness, the best return it offers, minus the mismatch of the policy, the return lost to its own suboptimal decisions. Training changes only the mismatch, while changing the harness moves both terms. Comparing harnesses with a fixed model therefore mixes what each harness offers with how well that model uses it. In MiniWoB++, post-training makes a model's preferences specific to its training harness. Reversing the order of the observation fields leaves every optimal decision unchanged yet lowers a trained model's success from 85.0% to 19.2%. Conversely, every new harness that a fixed-model comparison rejects but that raises the ceiling outperforms the original agent after further training under it, and ceiling and mismatch measured beforehand predict these gains. For coding-agent harnesses on Terminal-Bench Lite, ceiling minus mismatch estimated with a stronger model as a stand-in for the optimum reproduces the accuracy ranking, while either term alone does not. Code is available at https://github.com/2020-qqtcg/HRP.

Image Difference Captioning (IDC) aims to generate natural language descriptions of subtle differences between image pairs, requiring both precise visual change localization and coherent semantic expression. Despite recent advancements, existing datasets often lack breadth and depth, limiting their applicability in complex and dynamic environments: (1) from a breadth perspective, current datasets are constrained to limited variations of objects in specific scenes, and (2) from a depth perspective, prior benchmarks often provide overly simplistic descriptions. To address these challenges, we introduce OmniDiff, a comprehensive dataset comprising 324 diverse scenarios-spanning real-world complex environments and 3D synthetic settings-with fine-grained human annotations averaging 60 words in length and covering 12 distinct change types. Building on this foundation, we propose M^3Diff, a MultiModal large language model enhanced by a plug-and-play Multi-scale Differential Perception (MDP) module. This module improves the model's ability to accurately identify and describe inter-image differences while maintaining the foundational model's generalization capabilities. With the addition of the OmniDiff dataset, M^3Diff achieves state-of-the-art performance across multiple benchmarks, including Spot-the-Diff, IEdit, CLEVR-Change, CLEVR-DC, and OmniDiff, demonstrating significant improvements in cross-scenario difference recognition accuracy compared to existing methods. The dataset, code, and models will be made publicly available to support further research.

The throughput gains of speculative decoding depend on the trade-off between acceptance length and execution latency. We ask when additional acceptance is worth the cost of obtaining it. We establish a comparison protocol spanning training, inference, and evaluation, remeasuring released checkpoints and retraining six drafter recipes on shared target-generated data across ten benchmarks and two target models. Alignment changes the picture: gaps that appear large in published comparisons shrink substantially under shared training and inference conditions. This narrowing motivates two deeper questions: how is acceptance improved at different costs, and when do these gains improve throughput? We find that acceptance can be improved through nearly free objective and context changes or more expensive token-to-token conditioning. DCore illustrates the former route as a strong baseline with little additional measured drafting overhead. The payoff is conditional: task uncertainty and serving conditions affect acceptance returns, execution costs, and the preferred drafter and budget. These findings inform future drafter design and deployment choices.

Context: The SZZ algorithm is the de facto standard for labeling bug fixing commits and finding inducing changes for defect prediction data. Recent research uncovered potential problems in different parts of the SZZ algorithm. Most defect prediction data sets provide only static code metrics as features, while research indicates that other features are also important. Objective: We provide an empirical analysis of the defect labels created with the SZZ algorithm and the impact of commonly used features on results. Method: We used a combination of manual validation and adopted or improved heuristics for the collection of defect data. We conducted an empirical study on 398 releases of 38 Apache projects. Results: We found that only half of the bug fixing commits determined by SZZ are actually bug fixing. If a six-month time frame is used in combination with SZZ to determine which bugs affect a release, one file is incorrectly labeled as defective for every file that is correctly labeled as defective. In addition, two defective files are missed. We also explored the impact of the relatively small set of features that are available in most defect prediction data sets, as there are multiple publications that indicate that, e.g., churn related features are important for defect prediction. We found that the difference of using more features is not significant. Conclusion: Problems with inaccurate defect labels are a severe threat to the validity of the state of the art of defect prediction. Small feature sets seem to be a less severe threat.

Kaplan et al. and Hoffmann et al. developed influential scaling laws for the optimal model size as a function of the compute budget, but these laws yield substantially different predictions. We explain the discrepancy by reproducing the Kaplan scaling law on two datasets (OpenWebText2 and RefinedWeb) and identifying three factors causing the difference: last layer computational cost, warmup duration, and scale-dependent optimizer tuning. With these factors corrected, we obtain excellent agreement with the Hoffmann et al. (i.e., "Chinchilla") scaling law. Counter to a hypothesis of Hoffmann et al., we find that careful learning rate decay is not essential for the validity of their scaling law. As a secondary result, we derive scaling laws for the optimal learning rate and batch size, finding that tuning the AdamW beta_2 parameter is essential at lower batch sizes.