2385 papers
This paper presents a sensitivity-guided framework for compressing Reservoir Computing accelerators that systematically balances quantization and pruning to significantly improve hardware efficiency and reduce power consumption on FPGAs while maintaining high model accuracy across various time-series tasks.
This paper reviews FPGA-based AI accelerators for deep learning, highlighting their advantages over ASICs and GPUs, detailing key hardware optimization techniques such as loop pipelining and quantization, and analyzing state-of-the-art designs to identify challenges for future innovations.
The paper introduces Zipage, an LLM inference engine utilizing Compressed PagedAttention to combine token-wise KV cache eviction with PagedAttention, achieving over 2.1 speedup in high-concurrency reasoning tasks while maintaining approximately 95% of the performance of full KV inference.
This paper presents a systematic layer-wise and block-wise sensitivity analysis of NVFP4 and MXFP4 quantization across three Qwen2.5 model scales, revealing that MLP up- and down-projection layers are the most sensitive components while sensitivity patterns vary by format and model depth rather than being confined to final blocks.
This paper introduces the Variable-Invariant Two-Dimensional State Space Model (VI 2D SSM) and its unified VI 2D Mamba architecture, which theoretically establish and implement a permutation-equivariant framework for multivariate time series that eliminates artificial variable ordering to achieve state-of-the-art performance with improved structural scalability.
The paper introduces HCAPO, a novel framework that enhances long-horizon LLM agents by leveraging hindsight reasoning to refine step-level Q-values and employing a multi-scale advantage mechanism to address sparse reward challenges, thereby significantly outperforming state-of-the-art methods like GRPO on benchmarks such as WebShop and ALFWorld.
This paper introduces **Turn**, a compiled, actor-based programming language that enhances agentic software by integrating LLM inference as a typed primitive with schema validation, confidence-based control flow, isolated actor contexts, capability-based identity, and compile-time schema absorption to enforce critical safety and state invariants at the language level.
This paper introduces a principled method to reduce -invariant functions on product spaces to -invariant functions on alone, where is the isotropy subgroup of , thereby enabling flexible Equivariant Neural Fields to handle arbitrary group actions and heterogeneous product spaces without structural constraints.
The paper introduces EDMFormer, a transformer model trained on a newly released dataset of 98 professionally annotated EDM tracks (EDM-98) to address the limitations of existing music segmentation methods by leveraging genre-specific energy, rhythm, and timbre features for improved structure detection in Electronic Dance Music.
This paper critiques the current limitations of alignment-focused safety cases for frontier AI by drawing on established methodologies from safety-critical industries to propose a more robust, defensible framework, illustrated through a case study on deceptive alignment and CBRN capabilities.
This paper proposes a multi-level meta-reinforcement learning framework that systematically compresses Markov decision processes into hierarchical structures with skill-based curriculum learning to decouple sub-tasks, reduce stochasticity, and enable efficient transfer of skills across different problems and levels.
This paper introduces a large language model (LLM) framework that automates the control and measurement of superconducting qubits by dynamically generating and invoking tools based on a knowledge base, thereby enabling rapid deployment of standard protocols and the flexible implementation of novel experimental procedures.
This paper introduces Test-Driven AI Agent Definition (TDAD), a methodology that compiles tool-using LLM agents from behavioral specifications by iteratively refining prompts against executable tests, thereby ensuring measurable behavioral compliance and robustness against silent regressions through mechanisms like hidden test splits and semantic mutation testing.
Scale-Plan is a scalable framework that leverages large language models to filter irrelevant perceptual information and construct compact, task-relevant representations from natural language instructions, thereby enabling efficient and reliable long-horizon planning for heterogeneous multi-robot teams while outperforming existing baselines on the new MAT2-THOR benchmark.
This paper empirically demonstrates that coverage-based retrieval metrics serve as reliable early indicators of information coverage in RAG-generated responses, particularly when retrieval objectives align with generation goals, across diverse text and multimodal benchmarks.
This paper introduces Fish Audio S2, an open-source text-to-speech system that leverages a multi-stage training pipeline to enable multi-speaker, multi-turn generation with natural-language instruction following, while providing production-ready weights and an efficient SGLang-based inference engine.
This paper introduces GenGNN, a modular message-passing framework that demonstrates expressive neural backbones like transformers are not strictly necessary for discrete graph generation, as diffusion models using GenGNN achieve competitive validity and superior inference speed on various datasets.
MASEval introduces a framework-agnostic library that shifts multi-agent evaluation from a model-centric to a system-centric approach, demonstrating through extensive experiments that implementation decisions regarding topology and orchestration impact performance as significantly as model selection.
The paper presents MuCTaL, a lightweight multi-cancer tumor localization framework trained on four cancer types that achieves high performance on training data and demonstrates generalization to unseen tumor types, enabling scalable deployment for digital pathology applications.
This paper introduces the LLM Delegate Protocol (LDP), an AI-native communication framework that enhances multi-agent system efficiency and governance by exposing model identity and reasoning profiles as first-class primitives, demonstrating significant reductions in latency and token usage alongside improved security and recovery capabilities in experimental evaluations.