Papers915

Grounding and Enhancing Informativeness and Utility in Dataset Distillation

Shaobo Wang, Yantai Yang et al.Jan 29arXiv

This paper tackles dataset distillation by giving a clear, math-backed way to keep only the most useful bits of data, so models can learn well from far fewer images.

#dataset distillation#data condensation#Shapley value

Less Noise, More Voice: Reinforcement Learning for Reasoning via Instruction Purification

Beginner

Yiju Guo, Tianyi Hu et al.Jan 29arXiv

This paper shows that many reasoning failures in AI are caused by just a few distracting words in the prompt, not because the problems are too hard.

#LENS#Interference Tokens#Reinforcement Learning with Verifiable Rewards

Scaling Embeddings Outperforms Scaling Experts in Language Models

Intermediate

Hong Liu, Jiaqi Zhang et al.Jan 29arXiv

The paper shows that growing the embedding part of a language model (especially with n-grams) can beat adding more MoE experts once you pass a certain sparsity 'sweet spot.'

#N-gram Embedding#Mixture-of-Experts (MoE)#Embedding Scaling

Do Reasoning Models Enhance Embedding Models?

Intermediate

Wun Yu Chan, Shaojin Chen et al.Jan 29arXiv

The paper asks a simple question: if a language model becomes better at step-by-step reasoning (using RLVR), do its text embeddings also get better? The short answer is no.

#text embeddings#RLVR#contrastive learning

MAD: Modality-Adaptive Decoding for Mitigating Cross-Modal Hallucinations in Multimodal Large Language Models

Intermediate

Sangyun Chung, Se Yeon Kim et al.Jan 29arXiv

Multimodal AI models can mix up what they see and what they hear, making things up across senses; this is called cross-modal hallucination.

#multimodal large language models#cross-modal hallucination#contrastive decoding

CUA-Skill: Develop Skills for Computer Using Agent

Intermediate

Tianyi Chen, Yinheng Li et al.Jan 28arXiv

This paper builds a big, reusable library of computer skills so an AI can use Windows apps more like a careful human, not a clumsy robot.

#computer-using agents#desktop automation#skill library

Llama-3.1-FoundationAI-SecurityLLM-Reasoning-8B Technical Report

Beginner

Zhuoran Yang, Ed Li et al.Jan 28arXiv

This paper introduces Foundation-Sec-8B-Reasoning, a small (8 billion parameter) AI model that is trained to “think out loud” before answering cybersecurity questions.

#native reasoning#cybersecurity LLM#chain-of-thought

Thinking in Frames: How Visual Context and Test-Time Scaling Empower Video Reasoning

Intermediate

Chengzu Li, Zanyi Wang et al.Jan 28arXiv

This paper shows that making short videos can help AI plan and reason in pictures better than writing out steps in text.

#video reasoning#visual planning#test-time scaling

DeepSearchQA: Bridging the Comprehensiveness Gap for Deep Research Agents

Beginner

Nikita Gupta, Riju Chatterjee et al.Jan 28arXiv

DeepSearchQA is a new test with 900 real-world style questions that checks if AI agents can find complete lists of answers, not just one fact.

#DeepSearchQA#agentic information retrieval#systematic collation

Linear representations in language models can change dramatically over a conversation

Intermediate

Andrew Kyle Lampinen, Yuxuan Li et al.Jan 28arXiv

Language models store ideas along straight-line directions inside their brains (representations), like sliders for “truth” or “ethics.”

#linear representations#factuality#ethics

Idea2Story: An Automated Pipeline for Transforming Research Concepts into Complete Scientific Narratives

Intermediate

Tengyue Xu, Zhuoyang Qian et al.Jan 28arXiv

Idea2Story is a two-stage system that first studies many accepted research papers offline and then uses that knowledge online to turn a vague idea into a full scientific plan.

#autonomous scientific discovery#knowledge graph#method unit extraction

Training Reasoning Models on Saturated Problems via Failure-Prefix Conditioning

Intermediate

Minwu Kim, Safal Shrestha et al.Jan 28arXiv

When training smart language models with RL that use right-or-wrong rewards, learning can stall on 'saturated' problems that the model almost always solves.

#failure-prefix conditioning#RLVR#GRPO

15 16 17 18 19