Robuta

https://mediacentral.ucl.ac.uk/Play/112593 - Dr Clare Lyle - UCL NeuroAI Talk Series March 2024 Dr Clare Lyle (Google Deepmind) talks at the UCL Neuro AI Series March 2024 Talk title: Disentangling plasticity loss in neural networks clare lyletalk seriesdruclmarch https://clarelyle.com/posts/2021-12-18-offline.html Clare Lyle | The Importance of Making Mistakes in RL clare lyleimportancemakingmistakesrl https://clarelyle.com/posts/2018-11-10-grad-school.html Clare Lyle | Reflections on my first month in grad school clare lylefirst monthreflectionsgradschool https://clarelyle.com/index.html Clare Lyle | Home clare lyle https://clarelyle.com/posts/2026-02-15-sigmoid.html Clare Lyle | Sigmoids all the way down clare lyleway https://clarelyle.com/posts/2023-01-29-adam.html Clare Lyle | When do adaptive optimizers fail to generalize? clare lyleadaptiveoptimizersfail https://clarelyle.com/pubs.html Clare Lyle | Publications clare lylepublications https://clarelyle.com/posts/2025-06-30-plasticity-norm.html Clare Lyle | Does parameter norm growth cause networks to lose plasticity? clare lyleparameternormgrowthcause https://clarelyle.com/posts/2016-08-21-value.html Clare Lyle | What Work is Worth clare lyleworkworth https://clarelyle.com/posts/2020-10-30-bayesian.html Clare Lyle | A Bayesian Perspective on Training Speed and Model Selection clare lyletraining speedbayesianperspectivemodel https://clarelyle.com/posts/2023-12-27-plasticity.html Clare Lyle | Can you train a neural network forever? clare lyleneural networktrainforever https://clarelyle.com/posts/2025-10-18-sota.html Clare Lyle | Desperately Seeking SotA clare lyledesperatelyseekingsota https://clarelyle.com/posts/2026-03-22-ppo-reinforce.html Clare Lyle | Why is there no log in PPO? clare lylelogppo https://clarelyle.com/posts/2025-08-23-proximal.html Clare Lyle | A visual guide to proximal methods clare lylevisual guidemethods https://clarelyle.com/posts/2025-03-30-hidden-objectives.html Clare Lyle | Paper review: Auditing language models for hidden objectives clare lylelanguage modelspaperreviewauditing https://clarelyle.com/posts/2016-05-16-introtoinfinity.html Clare Lyle | A gentle introduction to infinities clare lylegentle introductioninfinities https://clarelyle.com/posts/2020-10-05-atoms.html Clare Lyle | Why I'm Learning Biology clare lylelearningbiology https://clarelyle.com/posts/2016-06-06-firstday.html Clare Lyle | First Week at Microsoft clare lylefirst weekmicrosoft https://clarelyle.com/posts/2023-08-27-llm-understanding.html Clare Lyle | Evaluating understanding in LLMs clare lyleevaluatingunderstandingllms https://clarelyle.com/posts/2017-05-20-settheory.html Clare Lyle | A semi-coherent summary of set theory clare lylesemicoherentsummaryset https://clarelyle.com/posts/2020-07-11-icml.html Clare Lyle | Causality, Generalization, and Reinforcement Learning clare lylecausalitygeneralizationreinforcementlearning https://clarelyle.com/posts/2016-05-25-stuff.html Clare Lyle | On Entering the Real World clare lyleenteringrealworld https://clarelyle.com/posts/2016-04-28-firstyearproofs.html Clare Lyle | A round-up of my favourite proofs from introductory math courses clare lyleroundfavouriteproofsintroductory https://clarelyle.com/posts/2025-09-06-plasticity-survey.html Clare Lyle | The state of plasticity in 2025 clare lylestateplasticity https://clarelyle.com/posts/2019-08-09-causality1.html Clare Lyle | Causality clare lylecausality https://clarelyle.com/posts/2025-11-02-bayesianism.html Clare Lyle | Thoughts on 'Rethinking the Role of Bayesianism in Modern AI' clare lylethoughtsrethinkingrolemodern https://clarelyle.com/posts/2026-02-26-not-a-loss.html Clare Lyle | On the policy gradient "loss" clare lylepolicygradientloss https://clarelyle.com/posts/2020-04-20-logs.html Clare Lyle | In praise of the logarithm clare lylepraise https://clarelyle.com/contact.html Clare Lyle | Contact clare lyle https://clarelyle.com/posts/2025-05-17-deriving-gauss-newton.html Clare Lyle | A post-rigorous derivation of the Gauss-Newton matrix clare lylepostrigorousderivationgauss