Transformers can learn broad families of tasks during pretraining and adapt to unseen tasks from a short prompt, but a rigorous understanding of this capability is limited. This paper studies how shared cross‑task structure influences the sample complexity of in‑context learning (ICL) by characterizing task‑space complexity through covering numbers, yielding a set of anchor functions that localize unseen tasks and predict responses. The authors construct a Transformer with Softmax attention to approximate this procedure and derive an error bound that separates the effects of pretraining tasks and prompt length, showing that once enough tasks are available the dependence on prompt length becomes dimension‑free.
By Zhongjie Shi, Rongjie Lai, Alexander Cloninger, Wenjing Liao
The paper argues that neural networks can exhibit modular organization even when individual neurons are not specialized. It introduces a nested hierarchy of modularity—unit-level, orthogonal, and linear—and shows that many vision, language, and RNN models lack unit-level modularity but still display clear functional modularity at the subspace level. The authors demonstrate that these subspaces can be causally validated through ablation and align with brain data, revealing distinct content/style or syntax/semantics separations.
By Shreya Saha, Benjamin Bergen, Meenakshi Khosla
arXiv:2609.38356v1 Announce Type: new
Abstract: Dynamical Systems Reconstruction (DSR) aims to infer models from observed time series that reproduce a system's qualitative long-term behavior. Continu...
By Sima Hashemi, Daniel Durstewitz, Georgia Koppe
arXiv:2603. 02462v2 Announce Type: replace-cross Abstract: A key challenge in developing unified neural solvers for combinatorial optimization (CO) is the efficient generalization of models from a given set of tasks to new tasks unseen during initial training.
By Semih Cant\"urk, Thomas Sabourin, Frederik Wenkel, Michael Perlmutter, Guy Wolf
NeuronSpark is a self‑attention‑free language model that uses input‑dependent spiking membrane dynamics, allowing each neuron’s decay rate, gain, and firing threshold to adapt to the current token. This selective spiking approach enables billion‑parameter pretraining from random initialization without relying on backbone transfer, conversion, or distillation, achieving competitive zero‑shot performance and a fast–slow hierarchy. On a 1.1B‑parameter, 2.1B‑token training run, NeuronSpark attains a 40.0 eight‑task zero‑shot average, surpasses comparable‑budget Pythia/Mamba checkpoints, and demonstrates significant energy and cycle reductions on FPGA implementations.
By Zhengzheng Tang
JIVEAdapter is a multi-task additive low-rank adapter that decomposes weight updates into a shared Joint component and a per-task Individual component, inspired by Joint and Individual Variation Explained (JIVE). It penalizes Individual structures to remain near-orthogonal to the Joint, ensuring interpretable separation of shared and task-specific signals, and allocates rank adaptively between shared and individual pools. The Joint component is learned once and frozen, then reused as a prior for new tasks, achieving competitive performance on GLUE and SuperGLUE with DeBERTaV3-base without additional modules.
By Sara Abdali, Pashmina Cameron