Teaching

MA-INF 4329 – Seminar Biological Intelligence (SoSe 2023)

Humans and other animals outperform artificial agents in various tasks and domains. This includes but is not limited to: learning and planning in unstructured domains; learning from sparse data, observation, and play; generalisation and transfer; causal reasoning; intuitive physics and psychology; language use; any time planning; continuous planning; spatial navigation; dynamic memory and active forgetting.

This seminar provides background on some of the underlying biological skills and computational theories that seek to explain them. We will discuss implications for designing and/or constraining artificial agents.

The first meeting takes place on 17 April (Monday) 16.00 (st) – 17.30 (Seminarraum 1.047, Friedrich-Hirzebruch-Allee 8, 1. OG). In this meeting the topics will be assigned.

If you are interested in participating in the seminar, please send an e-mail to Anja Menke and come to the first meeting.

If you would like to participate in the seminar but cannot attend the first meeting, please send us an email as well.

Updated course schedule:

8.5.2023One presentationHow BI differs from AI (visual recognition)

2nd paper
Gerhos et al.


Gerhos et al.
Generalisation in humans and deep neural networks

ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustness
https://proceedings.neurips.cc/paper/2018/hash/0937fb5864ed06ffb59ae5f9b5ed67a9-Abstract.html

https://arxiv.org/abs/1811.12231
15.5.2023One presentationBenchmarking AI against BI (cognitive psychology)Binz & SchulzUsing cognitive psychology to understand GPT-3https://doi.org/10.1073/pnas.2218523120
5.6.2023One presentationBenchmarking AI against BI (intuitive reasoning)Shu et al.AGENT: A Benchmark for Core Psychological Reasoninghttp://proceedings.mlr.press/v139/shu21a.html
12.6.2023Two presentationsAutomatising BI benchmarks (spatial navigation)
Incorporating BI into AI (intuitive reasoning)
Devlin et al.


Tsividis et al.
Navigation Turing Test (NTT): Learning to Evaluate Human-Like Navigation

Human-Level Reinforcement Learning through Theory-Based Modeling, Exploration, and Planning
https://arxiv.org/abs/2105.09637


https://arxiv.org/abs/2107.12544
19.6.2023Two presentationsBiological hardware constraints

BI case study: adaptive forgetting
Whittington & Bogasz

Bekinschtein et al.
Theories of Error Back-Propagation in the Brain

A retrieval-specific mechanism of adaptive forgetting in the mammalian brain
https://www.sciencedirect.com/science/article/pii/S1364661319300129

https://doi.org/10.1038/s41467-018-07128-7

26.6.2023
One presentationUnderstanding ANNs and BNNs 1

2nd or potential replacement paper
Chung & Abbott

Higgins et al.
Neural population geometry: An approach for understanding biological and artificial neural networks
Symmetry-Based Representations for Artificial and Biological General Intelligence
https://www.sciencedirect.com/science/article/pii/S0959438821001227

https://doi.org/10.3389/fncom.2022.836498