Papers7

#KL regularization

InfoNCE Induces Gaussian Distribution

Roy Betser, Eyal Gofer et al.Feb 27arXiv

The paper shows that when we train with the popular InfoNCE contrastive loss, the learned features start to behave like they come from a Gaussian (bell-shaped) distribution.

#InfoNCE#contrastive learning#Gaussian embeddings

GUI-Libra: Training Native GUI Agents to Reason and Act with Action-aware Supervision and Partially Verifiable RL

Intermediate

Rui Yang, Qianhui Wu et al.Feb 25arXiv

GUI-Libra is a training recipe that helps computer-using AI agents both think carefully and click precisely on screens.

#GUI agent#visual grounding#long-horizon navigation

ManCAR: Manifold-Constrained Latent Reasoning with Adaptive Test-Time Computation for Sequential Recommendation

Intermediate

Kun Yang, Yuxuan Zhu et al.Feb 23arXiv

ManCAR helps recommendation systems think step by step but keeps their thoughts on realistic paths using a map of how items connect.

#sequential recommendation#latent reasoning#interaction graph

Approximation of Log-Partition Function in Policy Mirror Descent Induces Implicit Regularization for LLM Post-Training

Intermediate

Zhenghao Xu, Qin Lu et al.Feb 5arXiv

The paper studies a simple way to train giant language models with reinforcement learning by replacing a hard-to-compute term (the log-partition function) with something easy: the mean reward.

#Policy Mirror Descent#KL regularization#chi-squared regularization

Latent Adversarial Regularization for Offline Preference Optimization

Intermediate

Enyi Jiang, Yibo Jacky Zhang et al.Jan 29arXiv

This paper introduces GANPO, a new way to train language models from human preferences by guiding the model using its hidden thoughts (latent space) instead of just its visible words (token space).

#GANPO#latent space regularization#offline preference optimization

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

Can LLMs Guide Their Own Exploration? Gradient-Guided Reinforcement Learning for LLM Reasoning

Intermediate

Zhenwen Liang, Sidi Lu et al.Dec 17arXiv

This paper teaches large language models (LLMs) to explore smarter by listening to their own gradients—the directions they would update—rather than chasing random variety.

#gradient-guided reinforcement learning#GRL#GRPO