Ga offline met de app Player FM !
Lecture | Sashank Varma | "Mathematical Concepts in Humans and Machine Learning Models"
Manage episode 377736952 series 2538953
Sashank Varma | Psychology and Interactive Computing, Georgia Institute of Technology
"Mathematical Concepts in Humans and Machine Learning Models"
The nature of mathematical concepts has long been a topic of philosophical debate. Recent theorizing in mathematical cognition has tended towards nativist accounts and postulations of built-in neural circuitry. In this talk, I consider whether this status quo is being challenged by the emergence of machine learning models capable of near-human levels of performance at predicting text and classifying images. Ongoing research in my lab is finding that these models induce latent representations of numerical and geometric concepts that are similar to those found in humans, for example, the mental number line. I will review several of these projects. I will also preview our future work, where we are moving beyond the cognitive alignment of machine learning models to evaluate their developmental alignment by training language models on developmentally calibrated corpora. The goal of this new work is first to model typical numerical development and then to perturb these typical models to shed light on developmental dyscalculia.
292 afleveringen
Manage episode 377736952 series 2538953
Sashank Varma | Psychology and Interactive Computing, Georgia Institute of Technology
"Mathematical Concepts in Humans and Machine Learning Models"
The nature of mathematical concepts has long been a topic of philosophical debate. Recent theorizing in mathematical cognition has tended towards nativist accounts and postulations of built-in neural circuitry. In this talk, I consider whether this status quo is being challenged by the emergence of machine learning models capable of near-human levels of performance at predicting text and classifying images. Ongoing research in my lab is finding that these models induce latent representations of numerical and geometric concepts that are similar to those found in humans, for example, the mental number line. I will review several of these projects. I will also preview our future work, where we are moving beyond the cognitive alignment of machine learning models to evaluate their developmental alignment by training language models on developmentally calibrated corpora. The goal of this new work is first to model typical numerical development and then to perturb these typical models to shed light on developmental dyscalculia.
292 afleveringen
همه قسمت ها
×Welkom op Player FM!
Player FM scant het web op podcasts van hoge kwaliteit waarvan u nu kunt genieten. Het is de beste podcast-app en werkt op Android, iPhone en internet. Aanmelden om abonnementen op verschillende apparaten te synchroniseren.