RP
Razvan Pascanu
cs.LGcs.AIstat.MLcs.NEcs.CLcs.CVNeural Networks, Computercs.DScs.ROartificial intelligence
On Valency
published · living versionsW_hbk3npz9·v1 · currentpublished
Preprints & journals
160 papers in the corpus · 2010–2026To Retain or to Adapt? Generalizing Continual Learning2607.05609v2 · Giulia Lanzillotta, Mandana Samiei, Doina Precup et al.2026 · 0 citationsTransactions on Machine Learning Research, 2026
Loss Smoothing for Stable Adaptation Under Distribution Shift2607.00634v1 · Darshan Patil, Ekaterina Lobacheva, Razvan Pascanu et al.2026 · 0 citationsarXiv
Overcoming Rank Collapse in Feedback Alignment2606.11123v2 · Gauthier Boeshertz, Razvan Pascanu, Claudia Clopath2026 · 0 citationsarXiv
Navigating Potholes with Geometry-Aware Sharpness Minimization2605.16134v1 · Simon Dufort-Labb'e, Mehrab Hamidi, Razvan Pascanu et al.2026 · 0 citationsarXiv
Revisiting Adam for Streaming Reinforcement Learning2605.06764v1 · Florin Gogianu, Adrian Catalin Lutu, Razvan Pascanu2026 · 0 citationsarXiv
Layerwise LQR for Geometry-Aware Optimization of Deep Networks2605.04230v1 · Simon Dufort-Labb'e, Pierre-Luc Bacon, Razvan Pascanu et al.2026 · 0 citationsarXiv
The Illusion of Stochasticity in LLMs2604.06543v2 · Xiangming Gu, Soham De, Michalis Titsias et al.2026 · 0 citationsarXiv
Understanding Performance Gap Between Parallel and Sequential Sampling in Large Reasoning Models2604.05868v1 · Xiangming Gu, Soham De, Larisa Markeeva et al.2026 · 0 citationsarXiv
Mining Generalizable Activation Functions2602.05688v1 · Alex Vitvitskyi, Michael Boratko, Matej Grcic et al.2026 · 0 citationsarXiv
Perplexity Cannot Always Tell Right from Wrong2601.22950v1 · Petar Velickovi'c, Federico Barbero, Christos Perivolaropoulos et al.2026 · 0 citationsarXiv
MS-SSM: A Multi-Scale State Space Model for Efficient Sequence Modeling2512.23824v1 · Mahdi Karami, Ali Behrouz, Peilin Zhong et al.2025 · 0 citationsarXiv
Fine-Tuned In-Context Learners for Efficient Adaptation2512.19879v1 · Jorg Bornschein, Clare Lyle, Yazhe Li et al.2025 · 0 citationsarXiv
Hadamard Product in Deep Learning: Introduction, Advances and Challenges.40232897 · Chrysos, Grigorios G, Wu, Yongtao, Pascanu, Razvan et al.2025 · 35 citationsIEEE transactions on pattern analysis and machine intelligence. 2025;47(8):6531-6549
What Can Grokking Teach Us About Learning Under Nonstationarity?2507.20057v1 · Clare Lyle, Gharda Sokar, Razvan Pascanu et al.2025 · 0 citationsarXiv
Optimizers Qualitatively Alter Solutions And We Should Leverage This2507.12224v1 · Razvan Pascanu, Clare Lyle, Ionut-Vlad Modoranu et al.2025 · 0 citationsarXiv
Filter Equivariant Functions: A symmetric account of length-general extrapolation on lists2507.08796v1 · Owen Lewis, Neil Ghani, Andrew Dudzik et al.2025 · 0 citationsarXiv
RAT: Bridging RNN Efficiency and Attention Accuracy via Chunk-based Sequence Modeling2507.04416v3 · Xiuying Wei, Anunay Yadav, Razvan Pascanu et al.2025 · 0 citationsarXiv
How Overconfidence in Initial Choices and Underconfidence Under Criticism Modulate Change of Mind in Large Language Models2507.03120v1 · Dharshan Kumaran, Stephen M Fleming, Larisa Markeeva et al.2025 · 1 citationarXiv
Meta-learning how to Share Credit among Macro-Actions2506.13690v1 · Ionel-Alexandru Hosu, Traian Rebedea, Razvan Pascanu2025 · 0 citationsarXiv
MesaNet: Sequence Modeling by Locally Optimal Test-Time Training2506.05233v2 · Johannes von Oswald, Nino Scherrer, Seijin Kobayashi et al.2025 · 0 citationsarXiv
Unpacking Softmax: How Temperature Drives Representation Collapse, Compression, and Generalization2506.01562v1 · Wojciech Masarczyk, Mateusz Ostaszewski, Tin Sum Cheng et al.2025 · 0 citationsarXiv
Plasticity as the Mirror of Empowerment2505.10361v2 · David Abel, Michael Bowling, Andr'e Barreto et al.2025 · 0 citationsarXiv
On the generalization of language models from in-context learning and finetuning: a controlled study2505.00661v3 · Andrew K. Lampinen, Arslan Chaudhry, Stephanie C.Y. Chan et al.2025 · 0 citationsarXiv
LLMs are Greedy Agents: Effects of RL Fine-tuning on Decision-Making Abilities2504.16078v1 · Thomas Schmied, Jorg Bornschein, Jordi Grau-Moya et al.2025 · 0 citationsarXiv
Lattice: Learning to Efficiently Compress the Memory2504.05646v2 · Mahdi Karami, Razvan Pascanu, Vahab Mirrokni2025 · 0 citationsarXiv
Why do LLMs attend to the first token?2504.02732v4 · Federico Barbero, 'Alvaro Arroyo, Xiangming Gu et al.2025 · 1 citationarXiv
NoProp: Training Neural Networks without Full Back-propagation or Full Forward-propagation2503.24322v2 · Qinyu Li, Yee Whye Teh, Razvan Pascanu2025 · 0 citationsarXiv
How do language models learn facts? Dynamics, curricula and hallucinations2503.21676v2 · Nicolas Zucchet, Jorg Bornschein, Stephanie Chan et al.2025 · 1 citationConference on Language Modeling (2025)
From Markov to Laplace: How Mamba In-Context Learns Markov Chains2502.10178v2 · Marco Bondaschi, Nived Rajaraman, Xiuying Wei et al.2025 · 0 citationsarXiv
Agency Is Frame-Dependent2502.04403v1 · David Abel, Andr'e Barreto, Michael Bowling et al.2025 · 0 citationsarXiv
Torque-Aware Momentum2412.18790v1 · Pranshu Malviya, Goncalo Mordido, Aristide Baratin et al.2024 · 0 citationsarXiv
TRecViT: A Recurrent Video Transformer2412.14294v2 · Viorica Patraucean, Xu Owen He, Joseph Heyward et al.2024 · 0 citationsarXiv
Non-Stationary Learning of Neural Networks with Automatic Soft Parameter Reset2411.04034v1 · Alexandre Galashov, Michalis K. Titsias, Andr'as Gyorgy et al.2024 · 1 citationNeurIPS 2024
A Large Recurrent Action Model: xLSTM enables Fast Inference for Robotics Tasks2410.22391v3 · Thomas Schmied, Thomas Adler, Vihang Patil et al.2024 · 2 citationsarXiv
Retrieval-Augmented Decision Transformer: External Memory for In-context RL2410.07071v3 · Thomas Schmied, Fabian Paischer, Vihang Patil et al.2024 · 0 citationsarXiv
Round and Round We Go! What makes Rotary Positional Encodings useful?2410.06205v3 · Federico Barbero, Alex Vitvitskyi, Christos Perivolaropoulos et al.2024 · 0 citationsarXiv
Softmax is not Enough (for Sharp Size Generalisation)2410.01104v3 · Petar Velickovi'c, Christos Perivolaropoulos, Federico Barbero et al.2024 · 1 citationarXiv
Switching between tasks can cause AI to lose the ability to learn.39169242 · Lyle, Clare, Pascanu, Razvan2024 · 2 citationsNature. 2024;632(8026):745-747
When can transformers compositionally generalize in-context?2407.12275v1 · Seijin Kobayashi, Simon Schug, Yassir Akram et al.2024 · 0 citationsarXiv
Investigating Low-Rank Training in Transformer Language Models: Efficiency and Scaling Analysis2407.09835v2 · Xiuying Wei, Skander Moalla, Razvan Pascanu et al.2024 · 1 citationarXiv
Normalization and effective learning rates in reinforcement learning2407.01800v1 · Clare Lyle, Zeyu Zheng, Khimya Khetarpal et al.2024 · 4 citationsarXiv
Building on Efficient Foundations: Effectively Training LLMs with Structured Feedforward Layers2406.16450v2 · Xiuying Wei, Skander Moalla, Razvan Pascanu et al.2024 · 0 citationsarXiv
Transformers meet Neural Algorithmic Reasoners2406.09308v1 · Wilfried Bounsi, Borja Ibarz, Andrew Dudzik et al.2024 · 3 citationsarXiv
State Soup: In-Context Skill Learning, Retrieval and Mixing2406.08423v1 · Maciej Pi'oro, Maciej Wo\lczyk, Razvan Pascanu et al.2024 · 0 citationsarXiv
Attention as a Hypernetwork2406.05816v4 · Simon Schug, Seijin Kobayashi, Yassir Akram et al.2024 · 0 citationsarXiv
Transformers need glasses! Information over-squashing in language tasks2406.04267v2 · Federico Barbero, Andrea Banino, Steven Kapturowski et al.2024 · 5 citationsarXiv
Deep Grokking: Would Deep Neural Networks Generalize Better?2405.19454v1 · Simin Fan, Razvan Pascanu, Martin Jaggi2024 · 1 citationarXiv
No Representation, No Trust: Connecting Representation, Collapse, and Trust Issues in PPO2405.00662v3 · Skander Moalla, Andrea Miele, Daniil Pyatko et al.2024 · 3 citationsarXiv
RecurrentGemma: Moving Past Transformers for Efficient Open Language Models2404.07839v2 · Aleksandar Botev, Soham De, Samuel L Smith et al.2024 · 5 citationsarXiv
Maxwell's Demon at Work: Efficient Pruning by Leveraging Saturation of Neurons2403.07688v2 · Simon Dufort-Labb'e, Pierluca D'Oro, Evgenii Nikishin et al.2024 · 0 citationsTransactions on Machine Learning Research 2835-8856 (2025)
Career total: 206 works. 160 are in this corpus.Showing the 50 most recent.
Profile built from the corpus for this byline.
Author records are still filling in while the Hub is in alpha. If this is your page, you'll be able to claim it soon. Spot a mistake? Tell us.