I will present a poster on collaborative work, led by Jai Bhagat, on SAEs for neural data at the Conference on the Mathematics of Neuroscience and AI in Rome.
I study how internal representations emerge in biological and artificial neural systems.
My background is in systems and computational neuroscience, where I completed a PhD studying how spatial and contextual information is encoded by biological neurons. I am now focusing on mechanistic interpretability and representation learning in artificial neural networks, supported by Coefficient Giving funding.
Latest news
I received funding from Coefficient Giving to support my career transition period into AI safety.
Current focus
- Mechanistic interpretability of deep learning models
- Compositional generalization in diffusion models
- Understanding feature representation in artificial neural networks
Currently based in London, UK. Email: saramolas18@gmail.com