AI Paper Atlas

Research map / Architectures and efficiency / Mixture of experts

Neural Thickets: Diverse Task Experts Are Dense Around Pretrained Weights

Yulu Gan, Phillip Isola

ICML 2026 · Architectures and efficiency › Mixture of experts · approach: Mixture-of-experts LoRA · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
scale model size from 0.5B to 32B (ablation)Qwen2.5 Gaussian-perturbed weights · GSM8Kfraction of perturbations matching or exceeding base: 0% -> 64%Improved
scale model size from 0.5B to 32B (ablation)Qwen2.5 Gaussian-perturbed weights · Countdownfraction of perturbations matching or exceeding base: 8% -> 60%Improved
increase model size (solution diversity) (ablation)Qwen2.5 family · seven tasks, 500 perturbationsSpectral DiscordanceImproved
ensemble top-50 perturbations instead of top-1 (K=50 vs K=1) (ablation)RandOpt · LLM post-training tasksaccuracyImproved
scale population size N (ablation)RandOpt · CountdownaccuracyImproved
increase selection ratio K (ablation)RandOpt with large N · CountdownaccuracyNo clear effect
add test-time majority vote over 50 samplesES · GSM8K, Qwen2.5-0.5Baccuracy: 42.6% -> 61.2%Improved
replace gradient/ES training with random guessing and ensemblingRandOpt vs GRPO · Countdown, OLMo3-7B-Instructaccuracy: 68.5 -> 85.0Improved
distill top-50 ensemble into a single model by SFTQwen2.5-1.5B-Instruct · GSM8Kaccuracy: 58.8 -> 74.9Improved
apply RandOpt to the language model of a VLMQwen2.5-VL-3B-Instruct · GQAaccuracy: 56.6 -> 69.0Improved
increase batch or group size to scale baseline parallelism in one step (ablation)GRPO and PPO · GSM8K, Qwen2.5-3B-Instructaccuracy: GRPO 83.5 -> 80.1No clear effect
pretrain on mixed signal types before random guessing (ablation)1D signal MLP · 1D signalsquality of nearby solutionsImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy