AINewsnow

Study Examines How Inference-Time Expert Deactivation Alters Routing and Weights in MoE Models

This story is from 2026-11-05. It is preserved in the archive; the latest stories are on the live feed.

A new arXiv paper analyzes expert deactivation at inference time, showing it changes both token routing and expert weights, with performance effects depending on how these changes are implemented.

Read the full story at arXiv cs.CL ↗

Timeline · 1 report

  1. 2026-11-05 05:00 · arXiv cs.CL
    Matching the Training-Time Weight Rule in Inference-Time Expert Deactivation

More stories

  1. Nous Research releases Arete-27B, cutting Qwen3.8-27B propaganda rate from 22% to 0.9% — AlphaSignal
  2. Hydra-PSIE expands 'Coeziune OM–IA' DOI registry to 38 published works — DEV Community — AI
  3. Man sues OpenAI, alleging ChatGPT helped his wife plot to poison him — Bloomberg AI
  4. AI stocks slide globally after Democrats' midterm wins spark tech rout — AP News
  5. 48-Hour Field Notes: Running Small AI Jobs on Free Model Access and Free Servers — DEV Community — AI
  6. Secrets leaking through coding-agent permission prompts, Apirio finds — DEV Community — AI
  7. Thinking Machines releases Inkling 2 Small, a 594B-parameter open-weight model, amid $5B funding round — Techmeme
  8. Ecosystem of tools grows around Claude Code and rival coding agents — DEV Community — AI

Get the daily brief of stories like this at 6:30 every morning →