The physics of Neural Networks

Tuesdays 10:30 - 12:00 - Seminar Room 4.18 (unless otherwise advertised)

Topics for talks

These topics are inspired by the chapters of the book “Complex and Adaptive Dynamical Systems” by Claudius Gros. It can be accessed on https://arxiv.org/pdf/0807.4838.pdf.

* The paper “Introduction to Random Boolean Networks” by Carlos Gershenson gives a nice overview of this topic.

The slides of the talks are accessible on http://zerocloud.detuning.org/s/jriwNkdjKzi3BHb.

Recommended research papers covering advanced topics:

• “Adaptive synchronization of neural networks with or without time-varying delay”, Jinde Cao, Jianquan Lu, Chaos, 2006.
• “Topological Evolution of Dynamical Networks: Global Criticality from Local Dynamics”, Stefan Bornholdt, Thimo Rohlf, Physical Review Letters, 2000.
• “Why deep and cheap learning works so well”, Henry W. Lin, Max Tegmark, and David Rolnick, Journal of Statistical Physics, 2017
• “Machine learning & artificial intelligence in the quantum domain”, Vedran Dunjko, Hans J. Briegel, Reports on Progress in Physics, 2018
• “Orchestrated reduction of quantum coherence in brain microtubules: A model for conciousness”, Stuart Hameroff, Roger Penrose, Mathematics and Computers in Simulation, 1996.
• “Importance of quantum decoherence in brain processes”, Max Tegmark, Physical Review E, 2000.
• “Conciousness as a State of Matter”, Max Tegmark, Chaos, Solitons & Fractals, 2015.
• “Modern Ergodic Theory”, Joel L. Lebowitz and Oliver Penrose, Physics Today, 1973

Literature

• “Complex and Adaptive Dynamical Systems”, Claudius Gros. In this seminar we will focus on the chapters 1-5, 7 and 8. The book can be accessed on https://arxiv.org/pdf/0807.4838.pdf.

Paper of the week

• “Nanophotonic media for artificial neural inference”, Erfan Khoram et al., Photonics Research, 2019 

Former papers of the week

• “Through-Wall Human Pose Estimation Using Radio Signals”, Mingmin Zhao et al., 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition.
• “Hybrid Forecasting of Chaotic Processes: Using Machine Learning in Conjunction with a Knowledge-Based Model”, Jaideep Pathak et al., Physical Review Letters, 2018.

Physicist’s Playground

• Code and train your own Neural Network for digit recognition:
  • MNIST training set comprising 60,000 handwritten digits^[1]:
    http://yann.lecun.com/exdb/mnist/index.html
  • Tutorial guiding through using this database:
    http://neuralnetworksanddeeplearning.com/chap1.html
• Neural Network simulation which allows you to train your own custom data set:
  https://lecture-demo.ira.uka.de/neural-network-demo/
• Conway’s Game of Life:
  • Cellular automaton with a set of simple updating rules
  • Initial state fully determines the evolution of cellular states
  • Online simulator allowing you to create your own initial states:
    https://bitstorm.org/gameoflife/

Figure: Structure moving on the grid of Conway’s Game of Life (upper). Evolving initial state on the surface of a knot (lower)^[2].

[1] The dataset is made available under a Creative Commons Attribution-Share Alike 3.0 license. No changes have been made to this dataset.

[2] Simulation done by Raphaelaugusto and made available under a Creative Commons Attribution-Share Alike 4.0 license. No changes have been made to this simulation.