BestHub
Sign in
Machine Learning Algorithms & Natural Language Processing
Machine Learning Algorithms & Natural Language Processing
Mar 26, 2026 · Artificial Intelligence

UniOD: A Single Model for Zero‑Training Cross‑Domain Anomaly Detection

UniOD introduces a universal outlier detection model that leverages historical labeled datasets to train one deep graph‑neural‑network‑based model, enabling plug‑and‑play anomaly detection on unseen domains without any retraining, and is backed by theoretical guarantees and extensive cross‑domain experiments.

UniODanomaly detectioncross-domain
0 likes · 10 min read
UniOD: A Single Model for Zero‑Training Cross‑Domain Anomaly Detection
Data Party THU
Data Party THU
Aug 27, 2025 · Artificial Intelligence

When Does Dot-Product Attention Switch from Positional to Semantic? A Phase Transition Theory

This paper presents a solvable low‑rank dot‑product attention model and, using high‑dimensional asymptotics and GAMP analysis, derives closed‑form characterizations of global optima that reveal a phase transition between positional and semantic attention mechanisms as sample complexity grows, with empirical validation against linear baselines.

GAMPdot-product attentionhigh-dimensional limit
0 likes · 10 min read
When Does Dot-Product Attention Switch from Positional to Semantic? A Phase Transition Theory
Data Party THU
Data Party THU
Aug 13, 2025 · Artificial Intelligence

How Dual Adaptivity Powers Universal Algorithms to Minimize Adaptive Regret

This article reviews the recent work by Zhou Zhihua’s team at Nanjing University on dual‑adaptivity universal algorithms for online convex optimization, introducing a meta‑expert framework, the UMA2 and UMA3 methods, and extending them to online composite optimization with strong adaptive‑regret guarantees.

Online Learningadaptive regretconvex optimization
0 likes · 10 min read
How Dual Adaptivity Powers Universal Algorithms to Minimize Adaptive Regret
Baobao Algorithm Notes
Baobao Algorithm Notes
Jun 3, 2025 · Artificial Intelligence

Can 1K Fine‑Tuning Replace 100K RL Steps? Insights from Re‑distillation Research

An extensive analysis shows that a 1K‑sample fine‑tuning stage can replicate the generalization gains of thousands of reinforcement‑learning steps, explains the compressibility of RL, introduces a sample‑effect theory, and demonstrates that re‑distillation and small‑scale SFT dramatically improve LLM performance.

Re-distillationSample Effectlarge language models
0 likes · 23 min read
Can 1K Fine‑Tuning Replace 100K RL Steps? Insights from Re‑distillation Research
AI Frontier Lectures
AI Frontier Lectures
May 25, 2025 · Artificial Intelligence

Can Alternating Generation‑Reduction Make LLMs Think Faster? Introducing PENCIL

The paper presents PENCIL, a novel alternating generation‑and‑erasure reasoning paradigm that achieves optimal space‑time complexity for chain‑of‑thought tasks, dramatically improves accuracy and efficiency on hard SAT, QBF, and Einstein puzzle benchmarks, and is provably Turing‑complete.

Pencilbenchmark resultschain of thought
0 likes · 12 min read
Can Alternating Generation‑Reduction Make LLMs Think Faster? Introducing PENCIL
NewBeeNLP
NewBeeNLP
May 13, 2024 · Artificial Intelligence

Why DPO Treats LLMs as Q‑Functions: A Deep Theoretical Dive

This article offers a detailed theoretical interpretation of the DPO algorithm, showing how large language models can be viewed as Q‑functions, unifying sequence‑wise and step‑wise decision perspectives, and discussing the resulting implications for reinforcement‑learning‑based alignment research.

DPOLLMQ-Function
0 likes · 14 min read
Why DPO Treats LLMs as Q‑Functions: A Deep Theoretical Dive