Lecture 11 - Homologies between brains and CNNs

Teacher: Rufin VanRullen (ANITI)

Very important: due to a strong response in the community, we are splitting the class in 2 sessions. This page is for the advanced session, appropriate for students with prior training in Maths, Engineering and/or Computer Science. If you are here by mistake, go to [the main/basic session page].

Lecture video

View the recorded lecture here (this will only be available for approximately 6 weeks after the course)

Contents

• What’s in a brain? Crash course in (visual) neuroscience:
  • Cortical Hierarchy
  • Receptive fields
  • Selectivities (features, object, classes)
  • Concept cells

• What’s in a CNN? Deepdream, visualization (explainability/interpretability) tools, examples…

• Brain/CNN comparisons:
  • RSA (representational similarity analysis): fMRI, MEG, single-units
  • Brainscore
  • Case study: CLIP-multimodal
• Other issues about the biological plausibility of Deep Learning:
  • Spikes
  • Adversarial attacks
  • Backprop
  • Attention/transformers
  • Recurrence…

---

(Back to Main Page)