Your personalresearch library

Discover papers, track what you're reading, rate and review.Build a library that grows with your curiosity.

Get Started Explore Papers

Currently Reading

Want to Read

All Papers

Publication Date

Newest Oldest

Average Rating

Highest Lowest

The astrocytic ensemble acts as a multiday trace to stabilize memory

Ken-ichi Dewa, Kodai Kaseda, Aoi Kuwahara, Hideaki Kubotera, Ayato Yamasaki, Natsumi Awata, Atsuko Komori, Mika A. Holtz, Atsushi Kasai, Henrik Skibbe, Norio Takata, Tatsushi Yokoyama, Makoto Tsuda, Genri Numata, Shun Nakamura, Eiki Takimoto, Masayuki Sakamoto, Minako Ito, Takahiro Masuda, Jun Nagai

Nature·2025·7 citations

Abstract Recalled memories become transiently labile and require stabilization 1–3 . The mechanism for stabilizing memories of survival-critical experiences, which are often emotionally salient and repeated, remains unclear 4 . Here we identify an astrocytic ensemble that is transcriptionally primed by emotional experience and functionally triggered by repeated experience to stabilize labile memory. Using a novel brain-wide Fos tagging and imaging method, we found that astrocytic Fos ensembles were preferentially recruited in regions with neuronal engrams 5 and were more widespread during fear recall than during conditioning. We established the induction mechanism of the astrocytic ensemble, which involves two steps: (1) an initial fear experience that induces day-long, slow astrocytic state changes with noradrenaline receptor upregulation; and (2) enhanced noradrenaline responses during recall, a repeated experience, enabling astrocytes to integrate coincident signals from local engrams and long-range noradrenergic projections, which induce secondary astrocytic state changes, including the upregulation of Fos and the neuromodulatory molecule IGFBP2. Pharmacological and genetic perturbation of the astrocytic ensemble signalling modulate engrams, and memory stability and precision. The astrocytic ensemble thus acts as a multiday trace in a subset of astrocytes after experience-dependent neural activity, which are eligible to capture future repeated experiences for stabilizing memories.

No ratings yet

View paper →

Challenges faced by visually impaired individuals from the perspective of faculty members: a phenomenological study

Yasir Ayed Alsamiri

Frontiers in Psychology·2025

Visually impaired students in higher education face significant academic, social, and psychological barriers that are often overlooked by faculty. This phenomenological study explored these challenges at Hail University, Saudi Arabia, by purposively sampling six faculty members from the Colleges of Law and Education to capture diverse disciplinary perspectives and ranks. Semi-structured interviews were audio-recorded, transcribed, and subjected to inductive thematic analysis with member checking and inter-rater reliability to ensure rigor. Three interrelated themes emerged: Academic Challenges—including inaccessible textbooks (lack of Braille and poorly tagged digital formats), rigid visual-based teaching methods, and absence of clear accommodation policies; Social Challenges—marked by peer ignorance, campus isolation, and exclusion from collaborative learning; and Psychological Challenges—manifested in faculty feelings of helplessness and student anxiety and low self-confidence stemming from repeated accessibility failures. Contrasting with Western models that centralize disability services, Hail University’s decentralized approach places the accommodation burden on individual faculty, underscoring the need for systemic reforms. We recommend establishing dedicated disability offices, mandating the concept of education training, launching campus-wide awareness campaigns, and integrating targeted mental-health support to foster equitable and sustainable learning environments.

★ 4.0 (1)

View paper →

PreviousPage 60 of 444Next

CurLL: A Developmental Framework to Evaluate Continual Learning in Language Models

Pavan Kalyan, Shubhra Mishra, Satya Lokam, Navin Goyal

arXiv·2025

We introduce a comprehensive continual learning dataset and benchmark (CurlL) grounded in human developmental trajectories from ages 5-10, enabling systematic and fine-grained assessment of models' ability to progressively acquire new skills. CurlL spans five developmental stages (0-4) covering ages 5-10, supported by a skill graph that breaks down broad skills into smaller abilities, concrete goals, and measurable indicators, while also capturing which abilities build on others. We generate a 23.4B-token synthetic dataset with controlled skill progression, vocabulary complexity, and format diversity, comprising paragraphs, comprehension-based QA (CQA), skill-testing QA (CSQA), and instruction-response (IR) pairs. Stage-wise token counts range from 2.12B to 6.78B tokens, supporting precise analysis of forgetting, forward transfer, and backward transfer. Using a 135M-parameter transformer trained under independent, joint, and sequential (continual) setups, we show trade-offs in skill retention and transfer efficiency. By mirroring human learning patterns and providing fine-grained control over skill dependencies, this work advances continual learning evaluations for language models.

No ratings yet

View paper →

Narrow Finetuning Leaves Clearly Readable Traces in Activation Differences

Julian Minder, Clément Dumas, Stewart Slocum, Helena Casademunt, Cameron Holmes, Robert West, Neel Nanda

arXiv·2025

Finetuning on narrow domains has become an essential tool to adapt Large Language Models (LLMs) to specific tasks and to create models with known unusual properties that are useful for research. We show that narrow finetuning creates strong biases in LLM activations that can be interpreted to understand the finetuning domain. These biases can be discovered using simple tools from model diffing - the study of differences between models before and after finetuning. In particular, analyzing activation differences on the first few tokens of random text and steering by adding this difference to the model activations produces text similar to the format and general content of the finetuning data. We demonstrate that these analyses contain crucial information by creating an LLM-based interpretability agent to understand the finetuning domain. With access to the bias, the agent performs significantly better compared to baseline agents using simple prompting. Our analysis spans synthetic document finetuning for false facts, emergent misalignment, subliminal learning, and taboo word guessing game models across different architectures (Gemma, LLaMA, Qwen) and scales (1B to 32B parameters). We suspect these biases reflect overfitting and find that mixing pretraining data into the finetuning corpus largely removes them, though residual risks may remain. Our work (1) demonstrates that narrowly finetuned models have salient traces of their training objective in their activations and suggests ways to improve how they are trained, (2) warns AI safety and interpretability researchers that the common practice of using such models as a proxy for studying broader finetuning (e.g., chat-tuning) might not be realistic, and (3) highlights the need for deeper investigation into the effects of narrow finetuning and development of truly realistic case studies for model-diffing, safety and interpretability research.

★ 4.0 (1)

View paper →

Searching for Stellar-Feedback-Driven Outflow Signatures: A Deep Dive into NGC 3741

Lexi N. Gault, Liese van Zee, Elizabeth A. K. Adams, James M. Wells, Laura Congreve Hunter, Kristen B. W. McQuinn, Roger E. Cohen, O. Grace Telford

arXiv·2025

Stellar feedback drives winds and outflows critical to the baryon cycles of low-mass galaxies whose shallow gravitational potential wells make them particularly susceptible to mass and metal loss through outflows. However, spatially resolved observations of stellar-feedback-driven outflows are limited due to their low-surface brightness and transient nature. We present the pilot of a larger multi-wavelength study searching for and quantifying stellar-feedback-driven winds and outflows on both spatially and globally resolved scales for a sample of 40 nearby low-mass galaxies. We search for outflow signatures in the star-forming dwarf galaxy NGC 3741 using new optical imaging and spectroscopy from the WIYN 3.5m telescope in conjunction with VLA 21cm observations and local star formation histories derived from resolved HST photometry. With this extensive dataset, we compare the neutral and ionized gas morphologies and kinematics, calculate mass-loading factors, and investigate spatial variations in the star formation history of NGC 3741. Though the galaxy is experiencing a burst in star formation, we find little evidence of strong outflows and calculate very low mass-loading factors. We suggest that, though star formation activity has increased dramatically in the central region of the galaxy over the last 40 Myr, the star formation rate is not high enough to produce a sufficient amount of high mass stars responsible for fueling outflows. Future analysis of the larger sample will allow us to explore how stellar feedback impacts mass loss on local scales, providing a deeper understanding of the interplay between stellar feedback and the interstellar medium in low-mass galaxies.

No ratings yet

View paper →

Robot Learning: A Tutorial

Francesco Capuano, Caroline Pascal, Adil Zouitine, Thomas Wolf, Michel Aractingi

arXiv·2025

Robot learning is at an inflection point, driven by rapid advancements in machine learning and the growing availability of large-scale robotics data. This shift from classical, model-based methods to data-driven, learning-based paradigms is unlocking unprecedented capabilities in autonomous systems. This tutorial navigates the landscape of modern robot learning, charting a course from the foundational principles of Reinforcement Learning and Behavioral Cloning to generalist, language-conditioned models capable of operating across diverse tasks and even robot embodiments. This work is intended as a guide for researchers and practitioners, and our goal is to equip the reader with the conceptual understanding and practical tools necessary to contribute to developments in robot learning, with ready-to-use examples implemented in $\texttt{lerobot}$.

No ratings yet

View paper →

Recontextualization Mitigates Specification Gaming Without Modifying the Specification

ariana_azarbal, Victor Gillioz, TurnTrout, cloud

AI Alignment Forum·2025

Recontextualization distills good behavior into a context which allows bad behavior. More specifically, recontextualization is a modification to RL which generates completions from prompts that discourage misbehavior, appends those completions to prompts that are more tolerant of misbehavior, and finally reinforces the model on the recontextualized instruction-completion data. Due to the data generation and training prompts differing in their attitude towards misbehavior, recontextualization builds resistance to misbehaviors that the training signal mistakenly reinforces. For example, suppose our reward signal does not robustly penalize deception. Recontextualization generates completions while discouraging deception and then creates training data by updating those completions' prompts to encourage deception. That simple tweak can prevent the model from becoming dishonest! Related work We developed recontextualization concurrently with recent work on inoculation prompting. Wichers et al. and Tan et al. find that when fine-tuning on data with an undesirable property, requesting that property in the train-time prompts prevents it from emerging under normal prompts at test-time. They use a fixed SFT dataset while we use an on-policy Reinforcement Learning procedure. Additionally, we not only request the bad property in train-time prompts, but we also prompt against the bad property in data generation. Recontextualization is a more general form of context distillation, which ...

No ratings yet

View paper →

Scaling Long-Horizon LLM Agent via Context-Folding

Weiwei Sun, Miao Lu, Zhan Ling, Kang Liu, Xuesong Yao, Yiming Yang, Jiecao Chen

arXiv·2025

Large language model (LLM) agents are fundamentally constrained by context length on long-horizon tasks. We introduce Context-Folding, a framework that empowers agents to actively manage their working context. An agent can procedurally branch into a sub-trajectory to handle a subtask and then fold it upon completion, collapsing the intermediate steps while retaining a concise summary of the outcome. To make this behavior learnable, we develop an end-to-end reinforcement learning framework FoldGRPO with specific process rewards to encourage effective task decomposition and context management. On complex long-horizon tasks (Deep Research and SWE), our folding agent matches or outperforms the ReAct baselines while using an active context 10$\times$ smaller and significantly outperforms models that rely on summarization-based context management.

No ratings yet

View paper →

Diffusion Transformers with Representation Autoencoders

Boyang Zheng, Nanye Ma, Shengbang Tong, Saining Xie

arXiv·2025

Latent generative modeling, where a pretrained autoencoder maps pixels into a latent space for the diffusion process, has become the standard strategy for Diffusion Transformers (DiT); however, the autoencoder component has barely evolved. Most DiTs continue to rely on the original VAE encoder, which introduces several limitations: outdated backbones that compromise architectural simplicity, low-dimensional latent spaces that restrict information capacity, and weak representations that result from purely reconstruction-based training and ultimately limit generative quality. In this work, we explore replacing the VAE with pretrained representation encoders (e.g., DINO, SigLIP, MAE) paired with trained decoders, forming what we term Representation Autoencoders (RAEs). These models provide both high-quality reconstructions and semantically rich latent spaces, while allowing for a scalable transformer-based architecture. Since these latent spaces are typically high-dimensional, a key challenge is enabling diffusion transformers to operate effectively within them. We analyze the sources of this difficulty, propose theoretically motivated solutions, and validate them empirically. Our approach achieves faster convergence without auxiliary representation alignment losses. Using a DiT variant equipped with a lightweight, wide DDT head, we achieve strong image generation results on ImageNet: 1.51 FID at 256x256 (no guidance) and 1.13 at both 256x256 and 512x512 (with guidance). RAE offers clear advantages and should be the new default for diffusion transformer training.

No ratings yet

View paper →

Your personalresearch library

All Papers

The astrocytic ensemble acts as a multiday trace to stabilize memory

Challenges faced by visually impaired individuals from the perspective of faculty members: a phenomenological study

CurLL: A Developmental Framework to Evaluate Continual Learning in Language Models

Narrow Finetuning Leaves Clearly Readable Traces in Activation Differences

Rosie Grant’s New Cookbook Compiles Recipes From Gravestones Around the World | Eater

Searching for Stellar-Feedback-Driven Outflow Signatures: A Deep Dive into NGC 3741

A Brief History & Ethos of the Digital Garden

Robot Learning: A Tutorial

Recontextualization Mitigates Specification Gaming Without Modifying the Specification

The echoes of childhood: How parental bonding and emotional trauma shape loneliness in young adults

Scaling Long-Horizon LLM Agent via Context-Folding

Diffusion Transformers with Representation Autoencoders