Papers1055

Specificity-aware reinforcement learning for fine-grained open-world classification

Samuele Angheben, Davide Berasi et al.Mar 3arXiv

This paper teaches AI to name things in pictures very specifically (like “golden retriever” instead of just “dog”) without making more mistakes.

#open-world classification#fine-grained recognition#large multimodal models

Chain of World: World Model Thinking in Latent Motion

Intermediate

Fuxiang Yang, Donglin Di et al.Mar 3arXiv

Robots learn better when they think about how things move over time, not by redrawing every pixel of a video.

#Vision-Language-Action#World Model#Latent Motion

BeyondSWE: Can Current Code Agent Survive Beyond Single-Repo Bug Fixing?

Intermediate

Guoxin Chen, Fanzhe Meng et al.Mar 3arXiv

BeyondSWE is a new benchmark that tests code agents on tougher, more real-life tasks than single-repo bug fixing.

#BeyondSWE#code agents#software engineering benchmark

NOVA: Sparse Control, Dense Synthesis for Pair-Free Video Editing

Intermediate

Tianlin Pan, Jiayi Dai et al.Mar 3arXiv

NOVA is a new video editor that lets you change a few key frames (sparse control) while it carefully keeps the original motion and background details (dense synthesis).

#video editing#pair-free training#sparse control

Next Embedding Prediction Makes World Models Stronger

Intermediate

George Bredis, Nikita Balagansky et al.Mar 3arXiv

NE-Dreamer is a model-based reinforcement learning agent that skips rebuilding pixels and instead learns by predicting the next step’s hidden features.

#model-based reinforcement learning#world models#next-embedding prediction

Heterogeneous Agent Collaborative Reinforcement Learning

Intermediate

Zhixia Zhang, Zixuan Huang et al.Mar 3arXiv

This paper introduces HACRL, a way for different kinds of AI agents to learn together during training but still work alone during use.

#HACRL#HACPO#heterogeneous agents

Track4World: Feedforward World-centric Dense 3D Tracking of All Pixels

Intermediate

Jiahao Lu, Jiayi Xu et al.Mar 3arXiv

Track4World is a fast, feedforward AI that can follow the 3D path of every pixel in a video using just one camera.

#dense 3D tracking#scene flow#2D-to-3D correlation

MemSifter: Offloading LLM Memory Retrieval via Outcome-Driven Proxy Reasoning

Intermediate

Jiejun Tan, Zhicheng Dou et al.Mar 3arXiv

MemSifter is a smart helper that picks the right memories for a big AI so the big AI doesn’t have to read everything.

#long-term memory#LLM retrieval#proxy model

ParEVO: Synthesizing Code for Irregular Data: High-Performance Parallelism through Agentic Evolution

Intermediate

Liu Yang, Zeyu Nie et al.Mar 3arXiv

ParEVO teaches AI to write fast, safe parallel code for messy, irregular data like big graphs and uneven trees.

#ParEVO#ParlayLib#irregular parallelism

PRISM: Pushing the Frontier of Deep Think via Process Reward Model-Guided Inference

Intermediate

Rituraj Sharma, Weiyuan Chen et al.Mar 3arXiv

PRISM is a new way to help AI think through hard problems by checking each step, not just the final answer.

#DEEPTHINK#Process Reward Model#step-level verification

HiFi-Inpaint: Towards High-Fidelity Reference-Based Inpainting for Generating Detail-Preserving Human-Product Images

Intermediate

Yichen Liu, Donghao Zhou et al.Mar 2arXiv

HiFi-Inpaint is a new AI method that fills a missing area in a photo of a person by inserting a specific product, while keeping tiny details like logos, textures, and small text crisp.

#reference-based inpainting#high-frequency map#Shared Enhancement Attention

Tool Verification for Test-Time Reinforcement Learning

Intermediate

Ruotong Liao, Nikolai Röhrich et al.Mar 2arXiv

The paper fixes a big flaw in test-time reinforcement learning (TTRL): when many wrong answers agree, the model rewards the mistake and gets stuck.

#test-time reinforcement learning#verification-weighted voting#tool verification

1 2 3 4 5