As Large Language Models (LLMs) gain agentic abilities, they will have to navigate complex multi-agent scenarios, interacting with human users and other agents in cooperative and competitive settings. This will require new reasoning skills, chief amongst them being theory of mind (ToM), or the ability to reason about the "mental" states of other agents. However, ToM and other multi-agent abilities in LLMs are poorly understood, since existing benchmarks suffer from narrow scope, data leakage, saturation, and lack of interactivity. We thus propose Decrypto, a game-based benchmark for multi-agent reasoning and ToM drawing inspiration from cognitive science, computational pragmatics and multi-agent reinforcement learning. It is designed to be as easy as possible in all other dimensions, eliminating confounding factors commonly found in other benchmarks. To our knowledge, it is also the first platform for designing interactive ToM experiments. We validate the benchmark design through comprehensive empirical evaluations of frontier LLMs, robustness studies, and human-AI cross-play experiments. We find that LLM game-playing abilities lag behind humans and simple word-embedding baselines. We then create variants of two classic cognitive science experiments within Decrypto to evaluate three key ToM abilities. Surprisingly, we find that state-of-the-art reasoning models are significantly worse at those tasks than their older counterparts. This demonstrates that Decrypto addresses a crucial gap in current reasoning and ToM evaluations, and paves the path towards better artificial agents.
Large language model-based machine learning (ML) agents have shown great promise in automating ML research. However, existing agents typically operate in isolation on a given research problem, without engaging with the broader research community, where human researchers often gain insights and contribute by sharing knowledge. To bridge this gap, we introduce MLE-Live, a live evaluation framework designed to assess an agent's ability to communicate with and leverage collective knowledge from a simulated Kaggle research community. Building on this framework, we propose CoMind, a novel agent that excels at exchanging insights and developing novel solutions within a community context. CoMind achieves state-of-the-art performance on MLE-Live and outperforms 79.2% human competitors on average across four ongoing Kaggle competitions. Our code is released at https://github.com/comind-ml/CoMind.
The global demand for sustainable protein sources has accelerated the need for intelligent tools that can rapidly process and synthesise domain-specific scientific knowledge. In this study, we present a proof-of-concept multi-agent Artificial Intelligence (AI) framework designed to support sustainable protein production research, with an initial focus on microbial protein sources. Our Retrieval-Augmented Generation (RAG)-oriented system consists of two GPT-based LLM agents: (1) a literature search agent that retrieves relevant scientific literature on microbial protein production for a specified microbial strain, and (2) an information extraction agent that processes the retrieved content to extract relevant biological and chemical information. Two parallel methodologies, fine-tuning and prompt engineering, were explored for agent optimisation. Both methods demonstrated effectiveness at improving the performance of the information extraction agent in terms of transformer-based cosine similarity scores between obtained and ideal outputs. Mean cosine similarity scores were increased by up to 25%, while universally reaching mean scores of $\geq 0.89$ against ideal output text. Fine-tuning overall improved the mean scores to a greater extent (consistently of $\geq 0.94$) compared to prompt engineering, although lower statistical uncertainties were observed with the latter approach. A user interface was developed and published for enabling the use of the multi-agent AI system, alongside preliminary exploration of additional chemical safety-based search capabilities
With the rapid advancement of deep learning, particularly through generative adversarial networks (GANs) and diffusion models (DMs), AI-generated images, or ``deepfakes", have become nearly indistinguishable from real ones. These images are widely shared across Online Social Networks (OSNs), raising concerns about their misuse. Existing deepfake detection methods overlook the ``block effects" introduced by compression in OSNs, which obscure deepfake artifacts, and primarily focus on raw images, rarely encountered in real-world scenarios. To address these challenges, we propose PLADA (Pay Less Attention to Deceptive Artifacts), a novel framework designed to tackle the lack of paired data and the ineffective use of compressed images. PLADA consists of two core modules: Block Effect Eraser (B2E), which uses a dual-stage attention mechanism to handle block effects, and Open Data Aggregation (ODA), which processes both paired and unpaired data to improve detection. Extensive experiments across 26 datasets demonstrate that PLADA achieves a remarkable balance in deepfake detection, outperforming SoTA methods in detecting deepfakes on OSNs, even with limited paired data and compression. More importantly, this work introduces the ``block effect" as a critical factor in deepfake detection, providing a robust solution for open-world scenarios. Our code is available at https://github.com/ManyiLee/PLADA.
The rapid advancement of AI, particularly large language models (LLMs), has raised significant concerns about the energy use and carbon emissions associated with model training and inference. However, existing tools for measuring and reporting such impacts are often fragmented, lacking systematic metric integration and offering limited support for correlation analysis among them. This paper presents WattsOnAI, a comprehensive software toolkit for the measurement, analysis, and visualization of energy use, power draw, hardware performance, and carbon emissions across AI workloads. By seamlessly integrating with existing AI frameworks, WattsOnAI offers standardized reports and exports fine-grained time-series data to support benchmarking and reproducibility in a lightweight manner. It further enables in-depth correlation analysis between hardware metrics and model performance and thus facilitates bottleneck identification and performance enhancement. By addressing critical limitations in existing tools, WattsOnAI encourages the research community to weigh environmental impact alongside raw performance of AI workloads and advances the shift toward more sustainable "Green AI" practices. The code is available at https://github.com/SusCom-Lab/WattsOnAI.
Industrial Non-Intrusive Load Monitoring (NILM) is limited by the scarcity of high-quality datasets and the complex variability of industrial energy consumption patterns. To address data scarcity and privacy issues, we introduce the Synthetic Industrial Dataset for Energy Disaggregation (SIDED), an open-source dataset generated using Digital Twin simulations. SIDED includes three types of industrial facilities across three different geographic locations, capturing diverse appliance behaviors, weather conditions, and load profiles. We also propose the Appliance-Modulated Data Augmentation (AMDA) method, a computationally efficient technique that enhances NILM model generalization by intelligently scaling appliance power contributions based on their relative impact. We show in experiments that NILM models trained with AMDA-augmented data significantly improve the disaggregation of energy consumption of complex industrial appliances like combined heat and power systems. Specifically, in our out-of-sample scenarios, models trained with AMDA achieved a Normalized Disaggregation Error of 0.093, outperforming models trained without data augmentation (0.451) and those trained with random data augmentation (0.290). Data distribution analyses confirm that AMDA effectively aligns training and test data distributions, enhancing model generalization.
Different base language model families, such as Llama and Qwen, exhibit divergent behaviors during post-training with reinforcement learning (RL), especially on reasoning-intensive tasks. What makes a base language model suitable for reinforcement learning? Gaining deeper insight into this question is essential for developing RL-scalable foundation models of the next generation. In this work, we investigate how mid-training strategies shape RL dynamics, focusing on two representative model families: Qwen and Llama. Our study reveals that (1) high-quality mathematical corpora, such as MegaMath-Web-Pro, significantly improve both base model and RL performance, while existing alternatives (e.g., FineMath-4plus) fail to do so; (2) further adding QA-style data, particularly long chain-of-thought (CoT) reasoning examples, enhances RL outcomes, and instruction data further unlocks this effect; (3) while long-CoT improves reasoning depth, it can also induce verbosity of model responses and unstability of RL training, underscoring the importance of data formatting; (4) scaling mid-training consistently leads to stronger downstream RL performance. Building on these insights, we introduce a two-stage mid-training strategy, Stable-then-Decay, in which base models are first trained on 200B tokens with a constant learning rate, followed by 20B tokens across three CoT-focused branches with learning rate decay. This yields OctoThinker, a family of models demonstrating strong RL compatibility and closing the performance gap with more RL-friendly model families, i.e., Qwen. We hope our work will help shape pre-training strategies for foundation models in the RL era. To support further research, we release our open-source models along with a curated math reasoning-intensive corpus of over 70 billion tokens (i.e., MegaMath-Web-Pro-Max).
Large Language Models (LLMs) exhibit remarkable code generation capabilities but falter when adapting to frequent updates in external library APIs. This critical limitation, stemming from reliance on outdated API knowledge from their training data, even with access to current documentation, impedes reliable code generation in dynamic environments. To tackle this issue, we propose ReCode (rule-based Reinforcement learning for Code Update), a novel framework that mimics human programmer adaptation to API changes. Specifically, we construct a dataset of approximately 2,000 data entries to train the LLMs to perform version migration based on updated information. Then, we introduce a modified string similarity metric for code evaluation as the reward for reinforcement learning. Our experiments demonstrate that ReCode substantially boosts LLMs' code generation performance in dynamic API scenarios, especially on the unseen CodeUpdateArena task. Crucially, compared to supervised fine-tuning, ReCode has less impact on LLMs' general code generation abilities. We apply ReCode on various LLMs and reinforcement learning algorithms (GRPO and DAPO), all achieving consistent improvements. Notably, after training, Qwen2.5-Coder-7B outperforms that of the 32B parameter code instruction-tuned model and the reasoning model with the same architecture. Code is available at https://github.com/zjunlp/ReCode.
We present LaplaceGNN, a novel self-supervised graph learning framework that bypasses the need for negative sampling by leveraging spectral bootstrapping techniques. Our method integrates Laplacian-based signals into the learning process, allowing the model to effectively capture rich structural representations without relying on contrastive objectives or handcrafted augmentations. By focusing on positive alignment, LaplaceGNN achieves linear scaling while offering a simpler, more efficient, self-supervised alternative for graph neural networks, applicable across diverse domains. Our contributions are twofold: we precompute spectral augmentations through max-min centrality-guided optimization, enabling rich structural supervision without relying on handcrafted augmentations, then we integrate an adversarial bootstrapped training scheme that further strengthens feature learning and robustness. Our extensive experiments on different benchmark datasets show that LaplaceGNN achieves superior performance compared to state-of-the-art self-supervised graph methods, offering a promising direction for efficiently learning expressive graph representations.
Learning from multi-variate time-series with heterogeneous channel configurations remains a fundamental challenge for deep neural networks (DNNs), particularly in clinical domains such as intracranial electroencephalography (iEEG), where channel setups vary widely across subjects. In this work, we introduce multi-variate parallel attention (MVPA), a novel self-attention mechanism that disentangles content, temporal, and spatial attention, enabling flexible, generalizable, and efficient modeling of time-series data with varying channel counts and configurations. We use MVPA to build MVPFormer, a generative foundation model for human electrophysiology, trained to predict the evolution of iEEG signals across diverse subjects. To support this and future effort by the community, we release the SWEC iEEG dataset, the largest publicly available iEEG dataset to date, comprising nearly 10,000 hours of recordings from heterogeneous clinical sources. MVPFormer leverages MVPA to achieve strong generalization across subjects, demonstrating expert-level performance in seizure detection and outperforming state-of-the-art Transformer baselines on our SWEC, the MAYO, and the FNUSA dataset. We further validate MVPA on standard time-series forecasting and classification tasks, where it matches or exceeds existing attention-based models. Together, our contributions establish MVPA as a general-purpose attention mechanism for heterogeneous time-series and MVPFormer as the first open-source, open-weights, and open-data iEEG foundation model with state-of-the-art clinical performance. The code is available at https://github.com/IBM/multi-variate-parallel-transformer. The SWEC iEEG dataset is available at https://mb-neuro.medical-blocks.ch/public_access/databases/ieeg/swec_ieeg.
In machine learning, it is common to obtain multiple equally performing models for the same prediction task, e.g., when training neural networks with different random seeds. Model multiplicity (MM) is the situation which arises when these competing models differ in their predictions for the same input, for which ensembling is often employed to determine an aggregation of the outputs. Providing recourse recommendations via counterfactual explanations (CEs) under MM thus becomes complex, since the CE may not be valid across all models, i.e., the CEs are not robust under MM. In this work, we formalise the problem of providing recourse under MM, which we name recourse-aware ensembling (RAE). We propose the idea that under MM, CEs for each individual model should be considered alongside their predictions so that the aggregated prediction and recourse are decided in tandem. Centred around this intuition, we introduce six desirable properties for solutions to this problem. For solving RAE, we propose a novel argumentative ensembling method which guarantees the robustness of CEs under MM. Specifically, our method leverages computational argumentation to explicitly represent the conflicts between models and counterfactuals regarding prediction results and CE validity. It then uses argumentation semantics to resolve the conflicts and obtain the final solution, in a manner which is parametric to the chosen semantics. Our method also allows for the specification of preferences over the models under MM, allowing further customisation of the ensemble. In a comprehensive theoretical analysis, we characterise the behaviour of argumentative ensembling with four different argumentation semantics. We then empirically demonstrate the effectiveness of our approach in satisfying desirable properties with eight instantiations of our method. (Abstract is shortened for arXiv.)
We introduce a neural network approach for generating and customizing the trajectory of a robotic arm, that guarantees precision and repeatability. To highlight the potential of this novel method, we describe the design and implementation of the technique and show its application in an experimental setting of cognitive robotics. In this scenario, the NICO robot was characterized by the ability to point to specific points in space with precise linear movements, increasing the predictability of the robotic action during its interaction with humans. To achieve this goal, the neural network computes the forward kinematics of the robot arm. By integrating it with a generator of joint angles, another neural network was developed and trained on an artificial dataset created from suitable start and end poses of the robotic arm. Through the computation of angular velocities, the robot was characterized by its ability to perform the movement, and the quality of its action was evaluated in terms of shape and accuracy. Thanks to its broad applicability, our approach successfully generates precise trajectories that could be customized in their shape and adapted to different settings.
Quantized large language models (LLMs) have gained increasing attention and significance for enabling deployment in resource-constrained environments. However, emerging studies on a few calibration dataset-free quantization methods suggest that quantization may compromise the safety capabilities of LLMs, underscoring the urgent need for systematic safety evaluations and effective mitigation strategies. In this paper, we present comprehensive safety evaluations across various mainstream quantization techniques and diverse calibration datasets, utilizing widely accepted safety benchmarks. To address the identified safety vulnerabilities, we propose a quantization-aware safety patching framework, Q-resafe, to efficiently restore the safety capabilities of quantized LLMs while minimizing any adverse impact on utility. Extensive experimental results demonstrate that Q-resafe successfully re-aligns the safety of quantized LLMs with their pre-quantization counterparts, even under challenging evaluation scenarios. Project page is available at: https://github.com/Thecommonirin/Qresafe.
Federated learning (FL) is a decentralized collaborative machine learning (ML) technique. It provides a solution to the issues of isolated data islands and data privacy leakage in industrial ML practices. One major challenge in FL is handling the non-identical and independent distributed (non-IID) data. Current solutions either focus on constructing an all-powerful global model, or customizing personalized local models. Few of them can provide both a well-generalized global model and well-performed local models at the same time. Additionally, many FL solutions to the non-IID problem are benefited from introducing public datasets. However, this will also increase the risk of data leakage. To tackle the problems, we propose a novel data-free distillation framework, Federated Bidirectional Knowledge Distillation (FedBKD). Specifically, we train Generative Adversarial Networks (GAN) for synthetic data. During the GAN training, local models serve as discriminators and their parameters are frozen. The synthetic data is then used for bidirectional distillation between global and local models to achieve knowledge interactions so that performances for both sides are improved. We conduct extensive experiments on 4 benchmarks under different non-IID settings. The results show that FedBKD achieves SOTA performances in every case.
Structured pruning is a well-established technique for compressing neural networks, making it suitable for deployment in resource-limited edge devices. This paper presents an efficient Loss-Aware Automatic Selection of Structured Pruning Criteria (LAASP) for slimming and accelerating deep neural networks. The majority of pruning methodologies employ a sequential process consisting of three stages: 1) training, 2) pruning, and 3) fine-tuning, whereas the proposed pruning technique adopts a pruning-while-training approach that eliminates the first stage and integrates the second and third stages into a single cycle. The automatic selection of magnitude or similarity-based filter pruning criteria from a specified pool of criteria and the specific pruning layer at each pruning iteration is guided by the network's overall loss on a small subset of the training data. To mitigate the abrupt accuracy drop due to pruning, the network is retrained briefly after each reduction of a predefined number of floating-point operations (FLOPs). The optimal pruning rates for each layer in the network are automatically determined, eliminating the need for manual allocation of fixed or variable pruning rates for each layer. Experiments on the VGGNet and ResNet models on the CIFAR-10 and ImageNet benchmark datasets demonstrate the effectiveness of the proposed method. In particular, the ResNet56 and ResNet110 models on the CIFAR-10 dataset significantly improve the top-1 accuracy compared to state-of-the-art methods while reducing the network FLOPs by 52\%. Furthermore, the ResNet50 model on the ImageNet dataset reduces FLOPs by more than 42\% with a negligible 0.33\% drop in top-5 accuracy. The source code of this paper is publicly available online - https://github.com/ghimiredhikura/laasp.
Autoregressive (AR) models have achieved unified and strong performance across both visual understanding and image generation tasks. However, removing undesired concepts from AR models while maintaining overall generation quality remains an open challenge. In this paper, we propose Erasure Autoregressive Model (EAR), a fine-tuning method for effective and utility-preserving concept erasure in AR models. Specifically, we introduce Windowed Gradient Accumulation (WGA) strategy to align patch-level decoding with erasure objectives, and Thresholded Loss Masking (TLM) strategy to protect content unrelated to the target concept during fine-tuning. Furthermore, we propose a novel benchmark, Erase Concept Generator and Visual Filter (ECGVF), aim at provide a more rigorous and comprehensive foundation for evaluating concept erasure in AR models. Specifically, we first employ structured templates across diverse large language models (LLMs) to pre-generate a large-scale corpus of target-replacement concept prompt pairs. Subsequently, we generate images from these prompts and subject them to rigorous filtering via a visual classifier to ensure concept fidelity and alignment. Extensive experimental results conducted on the ECGVF benchmark with the AR model Janus-Pro demonstrate that EAR achieves marked improvements in both erasure effectiveness and model utility preservation. Code is available at: https://github.com/immc-lab/ear/
We introduce MIRAGE, a new benchmark for multimodal expert-level reasoning and decision-making in consultative interaction settings. Designed for the agriculture domain, MIRAGE captures the full complexity of expert consultations by combining natural user queries, expert-authored responses, and image-based context, offering a high-fidelity benchmark for evaluating models on grounded reasoning, clarification strategies, and long-form generation in a real-world, knowledge-intensive domain. Grounded in over 35,000 real user-expert interactions and curated through a carefully designed multi-step pipeline, MIRAGE spans diverse crop health, pest diagnosis, and crop management scenarios. The benchmark includes more than 7,000 unique biological entities, covering plant species, pests, and diseases, making it one of the most taxonomically diverse benchmarks available for vision-language models, grounded in the real world. Unlike existing benchmarks that rely on well-specified user inputs and closed-set taxonomies, MIRAGE features underspecified, context-rich scenarios with open-world settings, requiring models to infer latent knowledge gaps, handle rare entities, and either proactively guide the interaction or respond. Project Page: https://mirage-benchmark.github.io
Graph Neural Networks (GNNs) have shown remarkable performance in the healthcare domain. However, what remained challenging is quantifying the predictive uncertainty of GNNs, which is an important aspect of trustworthiness in clinical settings. While Bayesian and ensemble methods can be used to quantify uncertainty, they are computationally expensive. Additionally, the disagreement metric used by ensemble methods to compute uncertainty cannot capture the diversity of models in an ensemble network. In this paper, we propose a novel method, based on knowledge distillation, to quantify GNNs' uncertainty more efficiently and with higher precision. We apply self-distillation, where the same network serves as both the teacher and student models, thereby avoiding the need to train several networks independently. To ensure the impact of self-distillation, we develop an uncertainty metric that captures the diverse nature of the network by assigning different weights to each GNN classifier. We experimentally evaluate the precision, performance, and ability of our approach in distinguishing out-of-distribution data on two graph datasets: MIMIC-IV and Enzymes. The evaluation results demonstrate that the proposed method can effectively capture the predictive uncertainty of the model while having performance similar to that of the MC Dropout and ensemble methods. The code is publicly available at https://github.com/tailabTMU/UQ_GNN.
The classification of imbalanced data streams, which have unequal class distributions, is a key difficulty in machine learning, especially when dealing with multiple classes. While binary imbalanced data stream classification tasks have received considerable attention, only a few studies have focused on multi-class imbalanced data streams. Effectively managing the dynamic imbalance ratio is a key challenge in this domain. This study introduces a novel, robust, and resilient approach to address these challenges by integrating Locality Sensitive Hashing with Random Hyperplane Projections (LSH-RHP) into the Dynamic Ensemble Diversification (DynED) framework. To the best of our knowledge, we present the first application of LSH-RHP for undersampling in the context of imbalanced non-stationary data streams. The proposed method undersamples the majority classes by utilizing LSH-RHP, provides a balanced training set, and improves the ensemble's prediction performance. We conduct comprehensive experiments on 23 real-world and ten semi-synthetic datasets and compare LSH-DynED with 15 state-of-the-art methods. The results reveal that LSH-DynED outperforms other approaches in terms of both Kappa and mG-Mean effectiveness measures, demonstrating its capability in dealing with multi-class imbalanced non-stationary data streams. Notably, LSH-DynED performs well in large-scale, high-dimensional datasets with considerable class imbalances and demonstrates adaptation and robustness in real-world circumstances. To motivate our design, we review existing methods for imbalanced data streams, outline key challenges, and offer guidance for future work. For the reproducibility of our results, we have made our implementation available on GitHub.
Diffusion models are a powerful tool for probabilistic forecasting, yet most applications in high-dimensional chaotic systems predict future snapshots one-by-one. This common approach struggles to model complex temporal dependencies and fails to explicitly account for the progressive growth of uncertainty inherent to such systems. While rolling diffusion frameworks, which apply increasing noise to forecasts at longer lead times, have been proposed to address this, their integration with state-of-the-art, high-fidelity diffusion techniques remains a significant challenge. We tackle this problem by introducing Elucidated Rolling Diffusion Models (ERDM), the first framework to successfully unify a rolling forecast structure with the principled, performant design of Elucidated Diffusion Models (EDM). To do this, we adapt the core EDM components-its noise schedule, network preconditioning, and Heun sampler-to the rolling forecast setting. The success of this integration is driven by three key contributions: (i) a novel loss weighting scheme that focuses model capacity on the mid-range forecast horizons where determinism gives way to stochasticity; (ii) an efficient initialization strategy using a pre-trained EDM for the initial window; and (iii) a bespoke hybrid sequence architecture for robust spatiotemporal feature extraction under progressive denoising. On 2D Navier-Stokes simulations and ERA5 global weather forecasting at 1.5^\circ resolution, ERDM consistently outperforms key diffusion-based baselines, including conditional autoregressive EDM. ERDM offers a flexible and powerful general framework for tackling diffusion-based sequence generation problems where modeling escalating uncertainty is paramount. Code is available at: https://github.com/salvaRC/erdm