NEW AI Test-Time Training & Invariant Latent Topologies for In-Context Learning

"In this video, we introduce In-Place Test-Time Training (In-Place TTT), a framework that seamlessly endows LLMs with Test-Time Training ability. In-Place TTT treats the final projection matrix of the ubiquitous MLP blocks as its adaptable fast weights, enabling a “drop-in" enhancement for LLMs without costly retraining from scratch." See pre-print below. "Despite substantial domain differences, many reusable implicit logical structures are shared across domains. In order to effectively retrieve cross-domain examples for unseen domains under investigation, in this work, we further propose an effective retrieval method, called domain-invariant neurons-based retrieval (DIN-Retrieval)." See pre-print below. All rights w/ authors: Towards Effective In-context Cross-domain Knowledge Transfer via Domain-invariant-neurons-based Retrieval Jianzhi Yan1,2*, Zhiming Li2*, Le Liu1,2, Zike Yuan1,2, Shiwei Chen1,2, Youcheng Pan2, Buzhou Tang2†, Yang Xiang2†, Danny Dongning Sun2†, from 1 Harbin Institute of Technology, Shenzhen, China 2 Pengcheng Laboratory, Shenzhen, China arXiv:2604.05383 https://github.com/Leon221220/DIN-Retrieval In-Place Test-Time Training Guhao Feng1,2,⋆, Shengjie Luo1,⋆, Kai Hua1, Ge Zhang1, Di He2,†, Wenhao Huang1,†, Tianle Cai1 from 1 ByteDance Seed, 2 Peking University https://github.com/ByteDance-Seed/In-Place-TTT #aiexplained #airesearch #scienceexplained #aiagents