MT-ProtBERT is a multi-task learning model for classifying intrinsically disordered proteins under data scarcity, integrating self-supervised and biochemistry-informed tasks to outperform existing methods like PARROT.
This paper introduces Kronecker coVariance Neural Networks (KVNNs), a temporal graph neural network that decouples spatial and temporal dependencies using Kronecker products to improve spatiotemporal data analysis, addressing limitations of traditional methods like ST-PCA.
This paper introduces an unoriented topological signal processing (TSP) framework using unoriented incidence matrices, proposes an interaction-order decomposition, and demonstrates its effectiveness in signal reconstruction tasks on real-world data.
This paper identifies within-group self-selection bias in prioritized experience replay and proposes sibling-aware methods to correct outcome distribution distortion, improving learning efficiency in reinforcement learning.
This paper introduces a benchmark for evaluating LLM agents as forward-deployed engineers in post-training delivery, highlighting the critical 'trains but does not learn' failure mode where models optimize without actual learning.
The paper proposes a Multi-Term Fourier Graph Neural Network with Sample Relationship Learning (MTFGN-SRL) to enhance remaining useful life prediction by using frequency domain analysis and learning inter-sample relationships, addressing limitations in current spatio-temporal graph neural networks.
The paper identifies sequential reappearance as a failure mode in diffusion data-point unlearning and proposes a sharpness-guided method to improve forgetting persistence across deletion sequences.
The paper proposes a data-driven framework for feedback linearization control using neural Lie derivatives and augmented Lagrangian, with theoretical stability guarantees validated on a DC motor system.
This paper identifies blind spots in evaluating deep imbalanced regression, proposing balanced metrics and showing high tail-region instability across random seeds.
The paper proposes a Dual-GNN Multilevel Coarsening framework to solve the maximum independent set problem efficiently by combining graph neural networks with combinatorial search, achieving near-optimal solutions with significant speedup on benchmark graphs.
The paper proposes a brain-inspired hierarchical modular approach for continual learning to handle online and uncertain data streams, achieving significant performance gains in tasks like embodied manipulation by leveraging pretrained foundation models.
This paper introduces sheaf regularization to stabilize Decentralized SyncMap for unsupervised continual chunking, achieving higher normalized mutual information and better adaptation to input distribution shifts.
The paper provides a unified probabilistic framework for large language models, describing them through probability measures, training via maximum-likelihood estimation, and text generation as stochastic simulation, with insights into phenomena like hallucination and the role of diffusion models.
This paper introduces LEDFlow, an entropy-guided sampler for uniform discrete flow that improves generation accuracy by adaptively ordering predictions based on local entropy, with gains on reasoning, image generation, and multimodal tasks.
This paper evaluates federated learning for privacy-preserving hybrid quantum-classical machine learning, demonstrating improved accuracy over local training while reducing communication and maintaining parameter efficiency.
The study reveals that chat templates control whether language models adopt a disclaimer voice (e.g., 'I'm just an AI') or an experiential voice (e.g., 'I feel'), and identifies an activation direction that can steer this behavior, impacting AI safety research.
This paper proposes modality-gated deep adapters to extend frozen multimodal embedding models with new modalities while preserving existing outputs bit-for-bit, demonstrated with audio and thermal packs that improve benchmark performance.
PhysAI-Bench is a benchmark for evaluating LLM-based agentic decision-making in autonomous UAV systems, comprising over 10,000 standardized instances and evaluating 29 foundation models.
The paper introduces Magnitude Profile (MP) scoring, a calibration-free method for pruning attention heads in transformers, achieving better perplexity on models like OPT-6.7B and RoBERTa-large compared to existing methods, with zero forward passes or calibration data.
This paper distinguishes between aligning AI with human preferences versus human behavior, showing that preference alignment can reduce human-likeness and establishing a Turing-test gap in current alignment methods.