20
Self-EvolveRec: Self-Evolving Recommender Systems with LLM-based Directional Feedback
Self-EvolveRec uses large language models to provide qualitative feedback for automatically evolving and optimizing recommender system architectures.
Top stories in Research
20
Neural Harmonic Textures for High-Quality Primitive Based Neural Reconstruction
Neural Harmonic Textures are introduced to enhance the expressivity of primitive-based 3D reconstruction methods like Gaussian Splatting for high-frequency details.
20
Playing ZendoWorld: Challenging AI Agents on Active Visual Concept Induction
ZendoWorld is a new interactive environment designed to test how well AI agents can perceive visual inputs, form hypotheses, and design experiments to infer hidden rules.
20
Best-of-$N$ TTS Evaluation is Confounded by ASR Family Alignment
This paper demonstrates that evaluating zero-shot text-to-speech systems using automatic speech recognition verifiers is biased by the specific model family used for judgment.
20
Ideas Have Genomes: Benchmarking Scientific Lineage Reasoning and Lineage-Grounded Idea Generation
The newly introduced IdeaGene-Bench evaluates the ability of artificial intelligence systems to trace scientific lineages and generate new ideas based on existing research.
20
Towards Precision Therapy in Hepatocellular Carcinoma: A Clinical-Reasoning LLM for Risk Stratification and Treatment Guidance
Researchers developed HCC-STAR, a specialized large language model designed to analyze electronic medical records for liver cancer risk stratification and treatment recommendations.
20
MCP tool design: Practical approaches and tradeoffs
Amazon Web Services has published a guide detailing practical context engineering approaches to optimize Model Context Protocol tool design.
20
FSD-VLN: Fast-Slow Dual-System Modeling for Aerial Long-Horizon Vision-Language Navigation
A new dual-system framework called FSD-VLN improves aerial vision-language navigation by separating high-level semantic reasoning from low-latency flight control.
20
An Online Reference-Free Evaluation Framework for Flowchart Image-to-Code Generation
This paper introduces an online, reference-free evaluation framework to monitor the quality of flowchart image-to-code generation by vision-language models during inference.
20
Blind-Spots-Bench: Evaluating Blind Spots in Multimodal Models
A new evaluation suite called Blind-Spots-Bench has been developed to identify simple tasks that humans easily perform but multimodal AI models struggle to complete.
20
MasFACT: Continual Multi-Agent Topology Learning via Geometry-Aware Posterior Transfer
MasFACT is a continual learning framework designed to optimize and adapt communication topologies among multi-agent systems across evolving tasks.
20
AutoPersonas: A Multi-Timescale Loop Engine for Open-Ended Persona Evolution
The AutoPersonas framework addresses the issue of behavioral collapse in long-term persona agents by enabling open-ended evolution across multiple timescales.
20
InvestPhilBench: A Multi-Layer Benchmark for Evaluating Large Language Model Procedural Reasoning in Expert Investment Philosophy
Researchers have introduced InvestPhilBench, a multi-layer benchmark designed to evaluate how well large language models can reconstruct and apply expert investment decision frameworks.
20
Beware What You Autocomplete: Forensic Attribution of Backdoored Code Completions
Researchers have developed CodeTracer, a forensic framework designed to detect and trace malicious backdoor attacks in large language model code completion systems.
20
Track2Map: Online Deformable SLAM with Motion-Aware Pose Optimization in Robotic Surgery
Track2Map is a new real-time 3D Gaussian Splatting system designed to reconstruct deformable anatomy during robotic surgery without relying on precise camera trajectory priors.
20
3100 Opinions on Code Review in an AI World: Building Causal Theory from Practitioner Discourse
This study analyzes practitioner perspectives to build a causal theory regarding how AI coding agents affect the process and outcomes of code reviews.
20
Towards Efficient Large Language Model Serving: A Survey on System-Aware KV Cache Optimization
This survey paper categorizes and reviews recent system-level optimizations for key-value caches to improve the efficiency and throughput of large language model serving.
20
Human-like Object Grouping in Self-supervised Vision Transformers
A new behavioral benchmark evaluates how closely the emergent object segmentation properties of self-supervised vision transformers align with human visual perception.
20
Concretized Proposition Prompting Resolves Composition-Knowledge Dichotomy in Large Language Models
A prompting framework called Concretized Proposition Prompting has been developed to improve reasoning and knowledge retrieval in large language models, particularly for medical tasks.
20
Diffusion-GR2: Diffusion Generative Reasoning Re-ranker
The authors present Diffusion-GR2, a generative reasoning re-ranker that utilizes block-diffusion language models to accelerate inference speeds in recommendation systems.
20
Shift & Drift: A Zero-Shot Benchmark for Generalizable and Robust Autonomous Driving Motion Planning
Researchers have introduced Shift & Drift, a new benchmark designed to evaluate the robustness and generalization of autonomous driving motion planners under distribution shifts.
20
Can We Trust LLM's Logic? Quantifying Uncertainty, Coherence, and Robustness via a Graph-Based Framework
A new graph-based framework evaluates the logical consistency and uncertainty of large language model reasoning steps rather than just checking the final output.
20
Training and Evaluating Diffusion Policies with Long Context Lengths
The study benchmarks the performance of robotic diffusion policies as their observation context length is increased, addressing memory limitations in dexterous manipulation.
20
Validity of LLMs as data annotators: AMALIA on authority
A study evaluates the validity of Portugal's AMALIA language model when used as an automated data annotator for identifying moral authority.