Recent work has shown that RLHF is highly susceptible to backdoor attacks. However, existing methods often rely on rare tokens or fixed triggers, limiting their impact in realistic scenarios. In this work, we develop GREAT, a novel framework for crafting natural distributional backdoors in RLHF. Specifically, GREAT targets harmful response generation for a vulnerable user subpopulation featured by semantically violent requests paired with emotionally angry triggers. At the core of our framework is a trigger identification pipeline that operates in the model's latent embedding space, leveraging dimensionality reduction and clustering techniques to identify representative triggers. To enable this, we introduce a hierarchical and diversity-driven prompting strategy to construct Erinyes, a high-quality dataset of over 5,000 angry triggers curated from GPT-4.1. Our experiments show that GREAT significantly outperforms baselines in attack generalization to unseen triggers, while preserving standard utility and maintaining stealth under defenses.
Approximating solutions to partial differential equations (PDEs) is fundamental for the modeling of dynamical systems in science and engineering. Physics-informed neural networks (PINNs) are a recent machine learning-based approach, for which many properties and limitations remain unknown.
PINNs are widely accepted as less computationally efficient and accurate than traditional methods for solving PDEs, such as the finite element method. However, PINNs are commonly claimed to show promise in solving inverse problems and handling noisy or incomplete data. We compare the performance of PINNs in solving inverse problems with that of a traditional approach using the finite element method combined with a numerical optimizer. The models are tested on viscosity identification in 1D Burgers' equation and in 2D/3D Taylor-Green Vortex, in all cases with additive Gaussian noise applied to training and validation data.
We find that while PINNs may require less human effort and specialized knowledge, they are outperformed by the traditional approach. For example, for 2D Taylor-Green Vortex with {\sigma}=1 noise, the baseline has a mean prediction RMSE of 0.0013 compared to 0.01 for the best PINN variation. However, PINNs scale better than the baseline with the computational complexity of the problem. We identify failures during training to be addressed if the PINN performance on noisy inverse problems is to become more competitive.
Aleksandra Jekic, Afroditi Natsaridou, Signe Riemer-S{\o}rensen et al.· 0 citations
Symbolic regression has emerged as a powerful tool for artificial intelligence-driven scientific discovery by learning interpretable analytical expressions that reveal governing relationships directly from data. Existing methods, however, often rely on heuristic search, struggle to balance predictive accuracy with expression complexity in noisy settings, and offer limited characterization of symbolic uncertainty. Probabilistic approaches that address these challenges in a unified manner remain underexplored. We introduce a probabilistic symbolic regression framework that represents mathematical expressions as ensembles of symbolic trees. A regularizing prior over tree topology controls expression complexity, while an Occam's window-based posterior summary captures uncertainty across multiple plausible symbolic models. Given the limited existing theoretical treatment of symbolic regression, we develop posterior concentration guarantees when symbolic expressions approximate the underlying relationship arbitrarily well, with a near-parametric rate when an exact finite formula exists. Additionally, we establish a sharp oracle concentration result under symbolic misspecification. Comparisons of our proposed framework with state-of-the-art competitors demonstrate superior predictive accuracy, optimal symbolic complexity, and stable structural recovery when learning benchmark scientific equations, together with the identification of scientifically interpretable descriptor formulas in a challenging materials discovery application.
Somjit Roy, Pritam Dey, Bani K. Mallick et al.· 0 citations
A Federated Learning (FL) system collaboratively trains neural networks across devices and a server but is limited by significant on-device computation costs. Split Federated Learning (SFL) systems mitigate this by offloading a block of layers of the network from the device to a server. However, in doing so, it introduces large communication overheads due to frequent exchanges of intermediate activations and gradients between devices and the server and reduces model accuracy for non-IID data. We propose Ampere, a novel collaborative training system that simultaneously minimizes on-device computation and device-server communication while improving model accuracy. Unlike SFL, which uses a global loss by iterative end-to-end training, Ampere develops unidirectional inter-block training to sequentially train the device and server blocks with a local loss, eliminating the transfer of gradients. A lightweight auxiliary network generation method decouples training between the device and server, reducing frequent intermediate exchanges to a single transfer, which significantly reduces the communication overhead. Ampere mitigates the impact of data heterogeneity by consolidating activations generated by the trained device block to train the server block, in contrast to SFL, which trains on device-specific, non-IID activations. Extensive experiments on multiple CNNs and Transformers show that, compared to state-of-the-art SFL baseline systems, Ampere (i) improves model accuracy by up to 11.70 percentage points while training up to 18.6x faster, (ii) incurs up to 911x lower device-server communication overhead and up to 14.5x lower on-device computation, and (iii) reduces standard deviation of accuracy by 71.13% for various non-IID degrees highlighting superior performance when faced with heterogeneous data. Ampere is available from https://github.com/blessonvar/Ampere.
Zihan Zhang, Leon Wong, Blesson Varghese· 0 citations
Reach audiences
Advertise in front of researchers, engineers, and readers.
We consider the problem of learning models of spatial density functions, representing the steady-state density of mobile nodes moving on a two-dimensional terrain. Deriving such models can assist in network design and optimization problems, e.g., by accelerating the computation of the density function during a parameter sweep. We address the question of applicability of off-the-shelf mixture density network models and of, two varieties of, normalizing flows for the description of mobile node density over a disk. We introduce the use of M\"obius distributions to retain symmetric spatial relations. Our results indicate that mixtures of M\"obius distributions provide interpretable, parsimonious models for the studied steady state density distributions, that match or outperform the alternatives.
Clinical decision-making reflects diverse strategies shaped by regional patient populations and institutional protocols. However, most existing medical artificial intelligence (AI) models are trained on highly prevalent data patterns, which reinforces biases and fails to capture the breadth of clinical expertise. Inspired by the recent advances in Mixture of Experts (MoE), we propose a Mixture of Multicenter Experts (MoME) framework to address AI bias in the medical domain without requiring data sharing across institutions. MoME integrates specialized expertise from diverse clinical strategies to enhance model generalizability and adaptability across medical centers. We validate this framework using a multimodal target volume delineation model for prostate cancer radiotherapy. With few-shot training that combines imaging and clinical notes from each center, the model outperformed baselines, particularly in settings with high inter-center variability or limited data availability. Furthermore, MoME enables model customization to local clinical preferences without cross-institutional data exchange, making it especially suitable for resource-constrained settings while promoting broadly generalizable medical AI.
Yujin Oh, Sangjoon Park, Xiang Li et al.· 0 citations
Modeling speech variation is key to natural, expressive generation. Speaker embeddings are commonly used to condition personalized speech systems, but they are typically trained for speaker recognition, where intra-speaker variability is suppressed and inter-speaker separation is maximized. This objective leads to overly compact representations that may discard variations crucial for generation. We revisit this design choice and propose a sub-center modeling framework for speaker embeddings. Instead of a single prototype per speaker, we learn multiple sub-centers during discriminative training, allowing utterances to align with different prototypes. This strategy preserves structured intra-speaker variability while maintaining discriminability. In zero-shot voice conversion, our method improves intelligibility, increases pitch variability, achieves higher naturalness ratings, and retains strong speaker verification performance.
Ismail Rasim Ulgen, John H. L. Hansen, Carlos Busso et al.· 0 citations
Evolution Strategies (ES) have recently emerged as a memory-efficient post-training paradigm for LLM reasoning. However, the optimization behavior of ES remains understudied, making it hard to define its advantage scope compared to mainstream post-training paradigms (e.g., Group Relative Policy Optimization (GRPO)). By systematically investigating ES dynamics and mechanisms, this paper first identifies a performance advantage of ES over GRPO, theoretically and empirically showing that ES can lead to broader reasoning coverage, thereby better exploiting the reasoning capabilities of pretrained LLMs. Theoretically, we show that verifier-projected Jensen-Shannon diversity across the ES population is helpful to higher Pass@K performances. Empirically, unlike GRPO, which exhibits entropy collapse, ES improves Pass@1 while attaining higher Pass@K than GRPO. We further develop a sequential GRPO-ES training strategy that combines GRPO's strength in Pass@1 with ES's gains in Pass@K. Second, we find that despite substantial whole-model parameter drift, the task-performance gains of ES are only contributed to a sparse subset of larger-magnitude updates. This functional sparsity suggests that large parameter movement need not imply widespread functional change, and held-out evaluations further show that it does not necessarily lead to catastrophic forgetting. Finally, we study how hyperparameter design affects the effectiveness of ES, demonstrating that ES requires a smaller population size in a larger LLM. These findings position ES as a distinct reasoning post-training paradigm rather than a less effective, memory-efficient alternative to GRPO.
Yunpeng Ba, Zhi Zheng, Yue Xie et al.· 0 citations
Nitrogen rates for winter wheat are set before the season, under unknown prices and weather. The standard UK advice does not respond to prices, yet recent price swings moved the most profitable rate by tens of kilograms per hectare. Machine learning is often proposed as the fix. However, it is usually judged on prediction accuracy, and accurate prediction does not by itself make the recommended rate more profitable. Our insight is to score nitrogen advice directly by the profit it forgoes on measured yield response curves. We build a test bench on 892 such curves from two long running UK experiments, and sweep the nitrogen to grain price ratio to cover all price scenarios. On this bench, machine learning fails as a predictor. No model recovers the best rate within farm tolerance, and the benchmark noise shows none can. At normal prices, every model also loses to the standard advice on profit. The gain sits elsewhere. A simple correction step applied after the model cuts profit losses by a quarter, while better models and extra features give no gain. The same frozen correction cuts losses by 43% at the second site without any retraining. A hybrid of standard advice plus a damped correction removes bias and trims rare large losses. The same price sweep also prices emission cuts, at a cost comparable to current carbon prices. Machine learning therefore pays as a profit scored correction to standard advice, not as its replacement.
In recent years, transductive knowledge graph embedding models have been applied to tasks such as link prediction and query answering. Although knowledge graphs often contain rich numerical attributes, most embedding models neglect them, limiting their ability to represent real-world knowledge graphs with diverse information. In this work, we propose a neural regression model (LitEm) that enables transductive knowledge graph embedding models to predict numerical attributes within knowledge graphs. Experimental results demonstrate that LitEm achieves the best or second-best results on most attributes across FB15K-237, YAGO15K, DB15K, and Mutagenesis. Furthermore, we propose a co-training framework that jointly trains state-of-the-art transductive knowledge graph embedding models with LitEm, which improves link prediction performance mainly for bilinear models and simultaneously enables them to predict numerical attributes. In addition, the literal-awareness evaluation demonstrates that co-training helps models to encode and exploit attribute information in a "literal-aware'' manner, suggesting that the observed gains are not merely due to additional parameters. We publicly release our implementation at https://github.com/dice-group/dice-embeddings.
Rupesh Sapkota, Louis Mozart Kamdem Teyou, Moshood Yekini et al.· 0 citations
Every deployed sparse-attention or KV-cache-eviction rule keeps a subset of the keys, discards the rest, and renormalizes the attention weights over the kept set. Enumerating the exact best subset under that constraint on $168{,}192$ attention rows from five models shows that keeping the largest weights is already near-optimal, since the best subset closes only a median $2$ to $5\%$ of the remaining gap to full attention. If selection closes this little, published margins between eviction methods must come from elsewhere, so we measure the bytes each method holds. In the shared evaluation pipeline, the strongest query-agnostic methods hold the full cache because their per-head selections are stored as masks, and only ragged per-head storage frees that memory. Enforcing a nominal budget on one fixed selection costs $14$ to $62$ benchmark points. We trace an $87.6$-point retrieval margin to rankings computed while the question is visible. ContourKV, a training-free allocator built from the dropped-mass statistic, wins $93$ of $160$ paired comparisons against that state of the art and loses $22$ at the byte count of the budget-enforcing baselines, and it ties the strongest of them.
Latent world models plan by predicting how candidate actions advance learned latent dynamics. In self-predictive models, however, the encoder and predictor are optimized jointly and can co-adapt to latent transitions that are easy to predict but weakly constrained by the physical evolution of the scene. We introduce the cross-predictive JEPA (JEPA-x), which grounds latent dynamics in privileged physical trajectories. JEPA-x treats visual observations and physical states as corresponding views of the same action-conditioned trajectory, advances both through a shared predictor, and matches each prediction to the future representations of both modalities. This encourages the action-conditioned predictor to learn a common transition rule across the two views. Privileged physical state is used only during training, leaving a visual-only model at deployment. Empirical results show that JEPA-x reduces the rollout drift of a newly fitted predictor from $0.361$ to $0.104$ and increases mean control success from $53.6\%$ to $78.2\%$ on a multi-task suite spanning six evaluation subfamilies. We additionally show that direct physical-state regression improves decodability without improving forecastability or control, indicating that the benefit comes from shaping latent dynamics rather than merely encoding physical variables.
Kehan Wen, Ziming Li, Siyuan Luo et al.· 0 citations
A new machine-learning framework aims to improve the success rate of computational protein design while moving away from results that reproduce sequences found in nature.
MIT News · Artificial Intelligence· news.mit.eduAug 24, 2026
A new method for surgically removing training examples from a model reveals that as datasets grow, the link between what a model learns and what it produces dissolves.
