7036 papers
This paper introduces ContextBench, a benchmark for evaluating methods that generate fluent inputs to trigger specific latent features in language models, and demonstrates that enhanced Evolutionary Prompt Optimization variants achieve state-of-the-art performance in balancing elicitation strength with linguistic fluency.
The paper introduces Sysformer, a novel approach that safeguards frozen large language models by learning to adapt system prompts in the embedding space to significantly improve safety robustness against harmful inputs and jailbreaking attacks without requiring costly model fine-tuning.
The paper introduces SPoT, a novel tokenization strategy that positions Vision Transformer tokens continuously within images via an oracle-guided search to overcome grid-based limitations, significantly reducing the required token count while improving performance and interpretability.
This paper introduces Quantifying Cross-Attention Interaction (QCAI), a novel post-hoc explainable AI method that interprets cross-attention mechanisms in encoder-decoder transformers to improve the understanding of TCR-pMHC binding, achieving state-of-the-art performance on the newly established TCR-XAI benchmark of 274 experimentally determined structures.
This paper introduces DejaVu, a novel attack that exploits in-vehicular network vulnerabilities to induce subtle temporal misalignments between camera and LiDAR streams, thereby severely degrading multimodal fusion-based perception tasks like object detection and tracking in autonomous driving systems.
This paper proposes a novel student-teacher framework for autonomous driving that utilizes a graph-based multi-agent RL teacher to automatically generate diverse, adaptive traffic curricula, enabling a student agent to achieve superior robustness and balanced driving performance compared to traditional rule-based approaches.
The paper proposes the State Space Neural Operator (SS-NO), a parameter-efficient architecture that extends structured state space models with adaptive damping and learnable frequency modulation to achieve state-of-the-art performance in learning solution operators for diverse time-dependent partial differential equations.
This paper introduces Multivariate Fields of Experts, a new image prior framework that generalizes existing methods using multivariate potential functions to achieve fast, interpretable, and theoretically guaranteed convergence in various inverse problems, outperforming univariate models while approaching deep learning performance with significantly fewer parameters and data.
This paper provides a theoretical characterization of the Expectation-Maximization algorithm's behavior in overspecified two-component mixed linear regression, establishing that unbalanced initial mixing weights yield linear convergence and optimal statistical accuracy, whereas balanced initial weights result in sublinear convergence and degraded accuracy.
This paper introduces Kernel VICReg, a novel self-supervised learning framework that extends the VICReg objective into a Reproducing Kernel Hilbert Space to capture nonlinear dependencies and improve representation learning performance on datasets with complex geometric structures.
The paper introduces ScaleZero, a multi-task planning model that combines a Mixture-of-Experts architecture to mitigate gradient conflicts and an online Dynamic Parameter Scaling strategy to efficiently allocate capacity, achieving performance comparable to specialized single-task agents with significantly reduced sample complexity.
This paper demonstrates that deep neural network ensembles enable accurate, real-time quantum parameter estimation with well-calibrated uncertainty quantification and drift detection, offering a faster alternative to traditional Bayesian inference methods without sacrificing accuracy.
This paper introduces "Answer-Then-Check," a novel safety alignment method that enhances LLM robustness against jailbreak attacks by training models to generate direct answers internally and then critically evaluate their safety before responding, achieving superior protection with reduced over-refusal while maintaining general reasoning capabilities through the newly constructed 80K-sample ReSA dataset.
VEGA is an electric vehicle navigation system that combines a physics-informed neural operator for real-time vehicle parameter estimation with a Proximal Policy Optimization agent for efficient, charge-aware route and charging stop planning, demonstrating superior inference speed and generalization across international road networks compared to traditional energy-aware baselines.
This paper introduces the Tensor Atomic Cluster Expansion (TACE), a unified atomistic machine learning framework that employs irreducible Cartesian tensors to efficiently model both scalar and tensorial observables without complex angular-momentum coupling, demonstrating robust accuracy and scalability across diverse chemical systems and tasks.
This paper proposes C²Prompt, a novel federated continual learning method that mitigates temporal and spatial forgetting by introducing a local class distribution compensation mechanism and a class-aware prompt aggregation scheme to enhance class-wise knowledge coherence across distributed clients.
This paper proposes a computationally efficient dual-network architecture combining an auto-regressive U-Net and a CNN to predict time-dependent full-field damage evolution and key mechanical properties in concrete, thereby enabling insights into aggregate effects and optimizing mix designs for improved durability.
This paper introduces GPHDM, a novel framework that extends Gaussian Process Dynamical Models to hyperbolic manifolds to generate physically consistent, human-like robot motions by preserving the hierarchical taxonomy and temporal dynamics of movement.
This paper addresses the training-inference mismatch in diffusion language models caused by planner-based sampling strategies by deriving a new Planned Evidence Lower Bound (P-ELBO) and introducing Planner Aware Path Learning (PAPL), a simple training modification that aligns training with planned inference to achieve significant performance gains across protein, text, and code generation tasks.
The paper introduces Diffusion Alignment as Variational Expectation-Maximization (DAV), an iterative framework that alternates between test-time search for diverse, reward-aligned samples and model refinement to optimize diffusion models for downstream objectives while mitigating reward over-optimization and mode collapse.