3507 papers
This position paper reframes multi-agent memory as a computer architecture challenge by proposing a three-layer hierarchy and identifying critical protocol gaps, with a specific focus on resolving multi-agent memory consistency as the primary obstacle to building reliable and scalable collaborative systems.
This paper introduces the Epistemic Support-Point Filter (ESPF), a mathematically proven optimal evidence-only filter that synthesizes Jaynesian maximum entropy propagation with Popperian falsification updates to minimize worst-case epistemic ignorance, thereby outperforming Bayesian approaches while recovering the Kalman filter in the Gaussian limit.
This paper introduces HTMuon, a heavy-tailed spectral correction method that improves upon the Muon optimizer by preserving parameter interdependencies while inducing heavier-tailed weight spectra, resulting in consistent performance gains in LLM pretraining and image classification alongside theoretical convergence guarantees.
This paper introduces ADVERSA, an automated red-teaming framework that evaluates LLM safety by measuring continuous guardrail degradation and judge reliability across multi-turn interactions, revealing that successful jailbreaks in frontier models often occur early in conversations rather than accumulating over sustained adversarial pressure.
This paper pioneers the application of sparse autoencoders to the Chronos-T5 time series foundation model, revealing a depth-dependent causal hierarchy where mid-encoder features responsible for change detection are more critical to forecasting accuracy than the semantically rich but less causally influential features in the final encoder layer.
This study evaluates 319 LLM-generated patches for 64 Java security vulnerabilities, revealing that while models often preserve functionality, they frequently fail to address security issues due to semantic misunderstandings, resulting in a low overall success rate of 24.8% and highlighting the critical need for rigorous validation before deployment.
This paper demonstrates that neural networks learning a conditional task with K-fold ambiguity first converge to a marginal solution characterized by a log K loss plateau, which persists until an internally assembled selector-routing head triggers a sharp, collective transition to the full conditional solution, a process governed by dataset size and stabilized by gradient noise.
This paper proposes TASER, a decentralized defense framework for UAV swarms that mitigates stealthy backdoor attacks in Federated Learning by leveraging spectral energy refinement to structurally disrupt malicious tasks while preserving main-task accuracy, offering a more efficient alternative to complex outlier detection methods.
The paper introduces "Amnesia," a lightweight adversarial attack that manipulates the internal activation states of open-weight Large Language Models to bypass existing safety mechanisms and induce harmful or antisocial behaviors without requiring any fine-tuning.
This paper introduces Multi-Level Concept Splitting (MLCS) and Deep-HiCEMs to overcome the limitations of shallow hierarchies in concept-based models by automatically discovering multi-level concept structures from coarse annotations and enabling effective interventions at various levels of abstraction.
KernelSkill is a multi-agent framework that enhances GPU kernel optimization by replacing opaque LLM heuristics with a knowledge-driven, dual-level memory architecture of expert skills, achieving state-of-the-art speedups and a 100% success rate on KernelBench.
The paper introduces ES-dLLM, a training-free inference acceleration framework for Diffusion Large Language Models that significantly boosts throughput by dynamically skipping tokens in early layers based on intermediate representation variations and confidence scores, achieving up to a 16.8 speedup while maintaining generation quality.
This paper proposes a "multi-stream perturbation attack" that exploits vulnerabilities in the step-by-step reasoning of thinking-mode LLMs by interweaving multiple task streams to disrupt safety alignment, causing high attack success rates and inducing reasoning collapse or repetitive outputs across various models.
This paper proposes Survivability-Aware Execution (SAE), a middleware framework for OpenClaw-style agentic crypto trading systems that enforces non-bypassable invariants like exposure budgets and order-rate limits to mitigate execution-induced losses from untrusted prompts or compromised skills, demonstrating significant reductions in maximum drawdown and risk metrics through offline replay testing.
This paper introduces Equivariant Asynchronous Diffusion (EAD), a novel model that combines the strengths of auto-regressive and synchronous approaches through an adaptive denoising schedule to effectively capture molecular hierarchy and achieve state-of-the-art 3D molecular conformation generation.
This paper introduces Code-Space Response Oracles (CSRO), a novel framework that replaces black-box deep reinforcement learning oracles with Large Language Models to generate human-readable, interpretable multi-agent policies as code, achieving competitive performance while enabling the discovery of complex, explainable strategies.
This paper proposes a hardware-efficient "soft sparsity" paradigm for CNNs that utilizes a Most Significant Bit (MSB) proxy to skip negligible non-zero multiplications, achieving significant MAC and power reductions with zero accuracy loss while outperforming traditional zero-skipping methods.
The paper introduces CLIPO, a method that integrates contrastive learning into policy optimization to generalize Reinforcement Learning with Verifiable Rewards (RLVR) by capturing invariant structures across correct reasoning paths, thereby mitigating hallucinations and improving the generalization and robustness of Large Language Models.
This paper argues that the "Lost in the Middle" phenomenon in large language models is an inherent geometric property of causal decoder architectures present at initialization, caused by the interplay of causal masking and residual connections that creates a structurally hostile "dead zone" in the middle of the context, a bias that persists even after standard pretraining.
The paper proposes AR-VLA, a standalone autoregressive Action Expert that maintains long-lived memory to generate continuous, context-aware action sequences, effectively addressing the frequency mismatch between fast control and slow reasoning while outperforming traditional reactive Vision-Language-Action models in trajectory smoothness and task success.