Find a story
Search Spins
Search titles, summaries, and missing voices across published articles — press releases, announcements, and media coverage.
0 results for “pruning”
Correlation-Aware Structured Pruning for Large Language Models
Researchers propose a new structured pruning method for LLMs that models correlations between model units to improve accuracy-efficiency trade-offs during inference cost reduction.
Sep 22, 2026
Pruning LLMs Like a Physicist: Block Removal as an Ising Optimization Problem
Hugging Face researchers introduced a novel method to prune large language models by framing block removal as an Ising model optimization problem, aiming to improve efficiency while preserving performance.
Sep 21, 2026
Demystifying Entropy-based Selection for Chain-of-Thought Compression in Large Reasoning Models
A new arXiv preprint challenges the efficacy of entropy-based pruning for Chain-of-Thought compression, finding no advantage over random pruning across models and tasks, and showing token-level entropy selection works only on math benchmarks due to numeric token properties—not generalizable reasoning heuristics.
Aug 3, 2026
Loss-Aware Feature-Map Pruning in Convolutional Neural Networks Using Multi-Armed Bandits
A new research paper introduces a feature-map pruning method for CNNs using multi-armed bandit algorithms to selectively remove redundant convolutional channels while preserving model accuracy and reducing compute.
Jul 28, 2026
Structured Pruning of Large Language Models via Power Transformation and Sign-Preserving Score Aggregation with Adaptive Feature Retention
A new structured pruning method for large language models improves inference speed while preserving accuracy by solving distribution mismatch, sign loss, and outlier sensitivity in adapting unstructured pruning techniques.
Jul 10, 2026
Pruning RAG context down to what the answer actually needs
A Hacker News thread discusses techniques for reducing retrieval-augmented generation (RAG) context size to improve answer relevance and efficiency.
Published Jul 6, 2026 · Analyzed Jul 8, 2026
On the Utility and Factual Reliability of Pruned Mixture-of-Experts Models in the Biomedical Domain
A new arXiv preprint investigates how pruning Mixture-of-Experts (MoE) models affects factual reliability in biomedical AI, finding that moderate pruning preserves utility but increases hallucination risk at extreme ratios—and that reliability degrades sharply outside the trained domain.
Published Jul 3, 2026 · Analyzed Jul 6, 2026