跟进版本变化

The rise of Mean Field Games (MFGs) has fostered a growing family of algorithms designed to solve large-scale multi-agent systems. However, the field currently lacks a standardized evaluation protocol, forcing researchers to rely on bespoke, isolated, and often simplistic environments. This fragmentation makes it difficult to assess the robustness, generalization, and failure modes of emerging methods. To address this gap, we propose a comprehensive benchmark suite for MFGs, focusing on the discrete-time, discrete-space, stationary setting. We introduce a taxonomy of problem classes, ranging from no-interaction and monotone games to potential and dynamics-coupled games, and provide prototypical environments for each. Furthermore, we present MF-Garnets, a method for generating random MFG instances to facilitate rigorous statistical testing. We benchmark a variety of learning algorithms across these environments, including a novel population-based approach (MF-PSO). To our knowledge, this is the first benchmark for MFGs.

Before deployment, an AI system might be evaluated on thousands or even millions of inputs, but after deployment it may encounter billions, so its worst failures may not appear during evaluation. Jones et al. (2025) proposed forecasting how risky the riskiest deployment inputs will be by scoring how close the system comes to failure on each evaluation input and extrapolating from the highest scores. We show that the error of this forecast splits into a noise term, whose distribution is the same for every model up to scale, and terms that depend on the shape of the tail of the model's risk scores. These model-dependent terms can make the forecast under-predict badly, both in theory, whenever the evaluation set lacks a rare kind of input that deployment contains, and in practice: on real user prompts marked as toxic, the forecast under-predicted the probability that a response begins with an offensive word for all eleven language models we tested. Because the shape of this tail is a property of the model, we propose fine-tuning the model itself to make the forecast accurate, using the forecast error as the training loss. In a model organism of jailbreaking and in an RL gridworld with rare traps, our method reduced the held-out squared forecast error by factors of 14 to 60, lowered worst-case risk about as much as direct risk minimization (or more), and preserved the model's primary behavior. For settings in which the model should change as little as possible, we show that training with an alternative safety constraint and an added loss term improved the forecast with far smaller changes to the model. Our method also improved the forecast in RL from sampled returns, in language models of up to 8B parameters, and (with minimal-change training) on the toxic user prompts.

Context: Open-source Pre-Trained Models (PTMs) provide extensive resources for various Machine Learning (ML) tasks, yet these resources lack a classification tailored to Software Engineering (SE). Objective: To address this gap, we derive a taxonomy encompassing 146 SE tasks and apply an SE-oriented classification to PTMs in a popular open-source ML repository, Hugging Face (HF). Method: Our repository mining study followed a five-phase pipeline: (i) identification of SE tasks from the literature; (ii) collection of PTM data from the HF API, including model card descriptions and metadata, and the abstracts of the associated arXiv papers; (iii) text processing to ensure consistency; (iv) a two-phase validation of SE relevance, involving humans and LLM assistance, supported by five pilot studies with human annotators and a generalization test; and (v) data analysis. This process yielded a curated catalogue of 5,027 SE PTMs. Results: We find that most SE PTMs target code generation and coding, emphasizing implementation over early or late development stages. In terms of ML tasks, text generation dominates within SE PTMs. Notably, the number of SE PTMs has increased markedly since 2023 Q2, while benchmarking remains limited: only 5.55\% of all SE PTMs in our catalogue provide evaluation results. Conclusions: Our catalogue reveals documentation and transparency gaps, highlights SE activity imbalances, and supports automated SE scenarios, such as the selection of suitable PTMs.

Molecular representations such as fingerprints and learned embeddings play a central role in cheminformatics and small-molecule machine learning. They serve as the primary input for a diverse array of tasks, including clustering, dataset splitting, and virtual screening, yet their evaluation is often limited to downstream model performance on narrow benchmark tasks. We carried out a systematic investigation of the quality of molecular representations, finding that none of the state-of-the-art molecular embedding models capture graph structure, 3D shape, or ADMET endpoint similarity better than fingerprints or physicochemical descriptors. However, classical fingerprints also revealed their own material limitations. For example, we show that depending on the representation, between 225M and $\sim$4B molecules have non-unique representations in the 10.8B Enamine REAL database, with implications for virtual screening. Our findings provide an insight into the current limitations of molecular representations, outlining future research directions for the development of more robust encodings of small molecules.

Machine learning interatomic potentials (MLIPs) are commonly trained and evaluated by their ability to reproduce first-principles energies and forces. Practical value, however, is often determined by downstream processes such as structural relaxation. Much of MLIP literature is motivated by the expectation that improving force accuracy will produce correspondingly better downstream behavior. Although this expectation is natural, it may not hold in practice due to model inaccuracies. Here, we demonstrate that force accuracy and relaxation accuracy can be driven in opposite directions. Starting from pretrained MLIPs, we differentiate through relaxation trajectories and optimize the the final structures. This intervention produces increases in force-prediction errors while substantially reducing structural errors. Importantly, this represents a trade-off between two aspects of physical fidelity: the learned forces become less faithful to first-principles forces, while the resulting relaxed structures become more faithful to first-principles predictions. Rather than observing a discrepancy between local quality and downstream performance, we demonstrate that this discrepancy can be induced through optimization. Our findings suggest that MLIPs need not be optimized as approximations of first-principles forces, but can instead learn task-specific vector fields that guide relaxation trajectories toward more accurate final structures. By propagating supervision through complete relaxation trajectories, our framework captures dependencies between successive predictions that are absent when structure--force pairs are treated independently. More broadly, these results demonstrate that conventional force metrics can provide an incomplete assessment of downstream performance and highlights the importance of evaluating scientific machine-learning models within computational workflows.

Large Language Models (LLMs) are powerful but limited by static parametric knowledge that becomes outdated once pretraining ends. Knowledge editing addresses this problem by updating model behavior on target facts without full retraining. In particular, in-context knowledge editing has gained attention because it is training-free and readily applicable to black-box LLMs. Recent reinforcement learning (RL)-based approaches improve over fixed retrieval strategies by adapting prompt construction to the quantity-quality trade-off. Despite initial success, they fail to model the prompt as a structured entity under the distinct and often competing objectives of reliability, generality, and specificity. Previous methods largely optimize a single objective and make decisions over only part of the prompt construction process, thereby overlooking both the balance of different objectives and the global organization of demonstrations. We propose Multi-Objective In-context Knowledge Editing (MO-IKE), a multi-objective RL algorithm that formulates prompt construction for in-context knowledge editing as a Constrained Markov Decision Process. MO-IKE trains a dynamic retriever to optimize competing objectives in knowledge editing, enabling more balanced and globally coherent prompt construction. On Llama-3.2, MO-IKE improves edit success (reliability) from 85.0% to 92.0%, paraphrase consistency (generality) from 77% to 79%, while increasing retention rate (specificity) by 23.0% compared to prior RL-based methods.

Collegiate cross country programs set race schedules without large comparable datasets across courses and conditions. Using the comprehensive-era National Running Club Database (NRCD; 23,360 results; 7,056 athletes; 2023-2025), where course and weather coverage exceed 99%, we ask what public results can identify about seasonal improvement, measurement, and racing opportunity. Hierarchical models of Standardized times recover pooled within-season improvement of about -5.7 s/week for men and -5.2 for women, but reading those slopes as fitness requires residual meet difficulty not to track the calendar: crossed meet effects can erase the week slope, while repeating-venue effects recover a clearer men's estimate under a weaker assumption. After the opener, race-1-only forecasts of who will improve are near noise (held-out R^2 ~ 0). Weather standardization remains a useful convention (Converted Only overstates first-to-last gains by about 15-21 s relative to Standardized), and prospectively trained venue factors modestly help where prior course data exist. Successful programs race more and place higher at nationals, but that may not be why they win: the schedule-placement link is between programs and is entangled with roster size, not evidence that changing one team's race count raises nationals place.

Emergency management assistance programs, such as relief distribution, are essential for delivering necessary supplies to affected communities. However, these programs operate in a decentralized network of local centers that face uncertain local demand and supply dynamics, resulting in inconsistent avail- ability of local services. Redistribution of supplies among these local centers reduces these imbalances, but the centers often make decisions independently, with limited information and disrupted transportation. This study develops CoRe-MARL, a cooperative multi-agent reinforcement learning (MARL) framework, by formulating a decentralized partially observable Markov decision process (Dec-POMDP). We treat each center as an agent that learns a redistribution policy to improve the service in the worst-case region and reduce the service gap across regions while protecting network-wide service. We incorporate a recurrent network that captures evolving supply and demand dynamics without direct observation, while multi-agent proximal policy optimization (MAPPO) enables centralized training and decentralized execution (CTDE). We evaluate the framework in a simulated environment with diverse trajectories, where exact dynamics are not observed by actors and the MAPPO critic. We compare the recurrent MAPPO with the recurrent independent PPO (IPPO) and a local only heuristic, and find that MAPPO reduces the service gap across local centers and enhances service for the worst-served center while maintaining competitive network-wide service. The recurrent MAPPO also shows consistent performance across diverse trajectory patterns, demonstrating its ability to adapt to evolving dynamics. The findings demonstrate the capability of cooperative learning for decentralized redistribution and improving equitable service under uncertain and evolving dynamics.

For effective resource management, infrastructure planning, and sustainable tourism, it is really important to predict accurate visitors. This paper estimates the daily visitation of Nebraska lakes using spatial, basic temporal, and interpretable temporal features. Moreover, we evaluate the base models with multiple metrics. Using the best three models, our method will ensemble hybrid models and choose the best hybrid technique. Further, we investigate the impact of lag features and how the models work with different outlier detection techniques such as Interquartile Range (IQR) and anomaly detection. The models are evaluated using regression metrics (MAE, MSE, RMSE, MAPE, and R2). Experiment results show that, after removing anomalies, MAE, MSE, RMSE, and MAPE are reduced. Adding lag features also shows some positive effects on the results. Moreover, anomaly detection techniques perform better than IQR in improving performance while preserving data. The reported anomaly-based cross-validation results favored the voting ensemble over the compared base models on the evaluated metrics. In anomaly-based cross-validation, VOTE improved prediction errors compared with MLP (256,128,64); improvements varied in the single-split comparison. Specifically, the hybrid model reduced MAE by 9.82%, MSE by 3.12%, RMSE by 1.76%, and MAPE by 6.77%. In the reported cross-validation comparison, the best anomaly-based hybrid model had lower MAE, MSE, RMSE, and MAPE than the best IQR-based hybrid model by 9.81%, 34.30%, 18.99%, and 0.66%, respectively.

Generative models, from diffusion models to large language models, achieve remarkable performance but at a cost in training data orders of magnitude larger than what biological learners require. An alternative paradigm has emerged in which networks are trained to predict their own latent representations of related views or masked regions, as in data2vec and JEPA -- an idea related to predictive-coding accounts of the cortex. Despite strong empirical results, the theoretical understanding of these methods remains limited. Central questions include: by how much does latent prediction actually improve data efficiency? Is there a benefit to stacking such methods into multi-scale hierarchies? We answer both using as data a tractable probabilistic context-free grammar that captures the compositional structure of natural language and images. Such a grammar generates strings of visible tokens by recursively applying production rules along a tree of hidden symbols of depth $L$. For such data, supervised or token-level SSL require a number of samples exponential in $L$ to recover the latent tree; we prove that latent prediction achieves this with a number of samples constant in $L$, up to logarithmic factors. We confirm this bound with (i) a hierarchical clustering algorithm, (ii) an end-to-end neural network whose predictor-clusterer modules predict their own latents at each level via gradient descent, and (iii) the first sample-complexity analysis of data2vec, which we show implicitly performs hierarchical latent prediction. This suggests that explicit stacking such as H-JEPA is largely redundant.

One-step generative models construct a static generator through iterative training-time transport. Existing transport objectives primarily assess distributional motion, although a neural generator needs to realize the requested sample displacements jointly through shared parameter updates. The training-time construction raises the question: \emph{once training becomes the iterative process that constructs the final one-step map, what to optimize: the next distributional move, or the route by which the finite generator learns the final map?} To address the question, we introduce \textbf{T}raining \textbf{D}ynamics \textbf{A}ction (\textbf{TDAction}), which selects transport targets according to local shared-parameter realization cost while retaining a prescribed level of distributional progress. We formulate the cost as a soft-terminal control problem and derive a closed-form Batch Tangent Action-to-Go value that accounts for parameter effort and terminal mismatch. The criterion captures cross-sample interactions omitted by independent pairwise costs; under isotropic mobility, the criterion agrees with quadratic Euclidean assignment for deterministic balanced couplings. Randomized tangent probes provide a low-rank implementation that constructs shared detached targets without adding an inference-time trajectory. Controlled studies examine the relationship between generator geometry, transport selection, and realized local action. On ImageNet $256\times256$, TDAction attains an FID below $1.1$ without distillation.

Learning to generalise from limited data is a fundamental challenge for both artificial and biological systems. A common strategy is to extract reusable structure from abundant unlabelled data, enabling efficient adaptation to new tasks from limited labelled data. This two-stage paradigm is now standard in modern training pipelines, where pretraining is followed by fine-tuning or linear probing. We provide an analytical model of this process: structure extraction is formalized as principal component analysis on unlabelled data, and downstream learning as linear regression on a separate labelled dataset. In the high-dimensional regime, we derive exact expressions for training and generalisation error showcasing their dependence on representation dimensionality, unlabelled and labelled sample sizes, and task alignment. Our results show that pretrained representations strongly influence downstream generalisation, and we characterize the optimal representation size as a function of task parameters: with abundant pretraining data but scarce downstream data, maximally compressed representations are optimal, whereas with limited pretraining data, higher-dimensional representations generalise better. Furthermore, we establish an exact trade-off between pretraining and supervision, quantifying how much unlabelled data is required to replace a single labelled sample. Beyond our idealised model, we observe similar phenomenology in autoencoders and pretrained LLMs. Altogether, we highlight that optimising representation size is critical, giving conditions for when compression during pretraining improves generalisation.

Selecting an energy-efficient model for periodic sensing requires accounting for time between inferences. Hardware-aware model search typically measures inference alone, even when the application cares about energy per release. We present CREST (Cross-platform Runtime Evaluation and Search Tool), an open-source hardware-in-the-loop experimentation system that evaluates model and hardware configurations under a specified release schedule. Researchers can vary the objective, schedule, target and candidate proposer while reusing evaluation and replay. Records preserve comparison conditions and separate valid measurements from deployable candidates. Three case studies demonstrate the system. First, searches and controlled replay on STM32 N657 show how waiting changes model and clock selection. For one architecture, continuous inference favors 800 MHz over 300 MHz at 8.0 versus 17.1 mJ per inference, but additional waiting offsets 97% of this saving in a fitted decomposition. With STOP between releases, 300 MHz uses about 0.3-0.4% less board energy per release at 100 and 500 ms. At a fixed clock, the preferred architecture changes between release periods, which a model with shared waiting parameters cannot produce. Latency predicts continuous energy within a median 1.7% on held-out observations, yet most release-energy error remains in scheduled terms. In paired replay, one of six randomly selected near-latency pairs reverses its continuous-energy order by 5.8 mJ per release. Second, replay on three Arduino targets shows proxy-selected inertial-odometry candidates can fail to fit or use more energy than measured selections, with only 9 of 16 FLOPs-selected candidates fitting the most memory-constrained board. Third, an audio classification study reuses the evaluator with LLM-based candidate proposers, and cross-target replay moves a selected configuration from third to last in energy rank.