Perceptual decision making in a recurrent neural network

Type and duration:

Master project, duration flexible.

Context:

Every day in our life we have to make various decisions. This starts already early in the morning when the alarm goes off. Should I sleep another five minutes or should I get up? What do these decision processes look like in our brain? Studies on primates have been conducted to explain how two-choice decision making could happen in the brain. The animals were watching a screen with dots moving randomly with some of them following a coherent movement direction (either right or left). By moving the eyes to one of the two cardinal directions the monkey indicates the main movement direction of the dots. The outcome of these observations led to the design of a simple recurrent neural network mechanism called Wang's Winner-Take-All which could explain how two-choice decision making takes place in the primate brain [1]. In this project you will investigate the performance of a simplified version of Wang's Winner-Take-All network including the Spiking Elementary Motion Detector [2] on the neuromorphic platform SpiNNaker [3]. You will learn the basic principles of spiking neural networks and how to use them together with event-based cameras [4]. The outcome of this thesis focuses on a complete understanding how decision making takes place in Wang's Winner-Take-All. This will help us to further improve our understanding of the primate brain and our own brain. As an extension of your thesis you can implement the working network on the Neuromorphic Pan-Tilt Unit (NEPTUn) [5] and test its performance in a real world scenario.

Objectives:

• Understand how recurrent neural networks can perform decision making.
• Implement Wang's Winner-Take-All onto the SpiNNaker platform.
• Optimize and characterize Wang's Winner-Take-All for the random dot motion task.
• Execute a closed-loop experiment with the Neuromorphic-Pan-Tilt-Unit (NePTUn) [5] replicating experiments with Zebrafish [6] (optional)

Required skills:

python programming, strong interest in computational and cognitive neurosciences.

Helpful but optional: FPGA programming and SpiNNaker programming.

Contact person:

Thorben Schoepe (external at Forschungszentrum Jülich)

References:

1. Wong KF, Wang XJ. A recurrent network mechanism of time integration in perceptual decisions. J Neurosci. 2006 Jan 25;26(4):1314-28. doi: 10.1523/JNEUROSCI.3733-05.2006. PMID: 16436619; PMCID: PMC6674568.
2. Milde, Moritz & Bertrand, Olivier & Ramachandran, H. & Egelhaaf, Martin & Chicca, Elisabetta. (2018). Spiking Elementary Motion Detector in Neuromorphic Systems. Neural Computation. 30. 1-34. 10.1162/neco_a_01112.
3. E. Painkras et al., "SpiNNaker: A 1-W 18-Core System-on-Chip for Massively-Parallel Neural Network Simulation," in IEEE Journal of Solid-State Circuits, vol. 48, no. 8, pp. 1943-1953, Aug. 2013, doi: 10.1109/JSSC.2013.2259038.
4. Lichtsteiner, P., Posch, C., & Delbruck, T. (2006). A 128 128 120 dB 15 s Latency Asynchronous Temporal Contrast Vision Sensor.
5. T. Schoepe, D. Gutierrez-Galan, J. P. Dominguez-Morales, A. Jimenez-Fernandez, A. Linares-Barranco and E. Chicca, "Live Demonstration: Neuromorphic Sensory Integration for Combining Sound Source Localization and Collision Avoidance," 2020 IEEE International Symposium on Circuits and Systems (ISCAS), Seville, Spain, 2020, pp. 1-1, doi: 10.1109/ISCAS45731.2020.9181257.
6. Naumann EA, Fitzgerald JE, Dunn TW, Rihel J, Sompolinsky H, Engert F. From Whole-Brain Data to Functional Circuit Models: The Zebrafish Optomotor Response. Cell. 2016 Nov 3;167(4):947-960.e20. doi: 10.1016/j.cell.2016.10.019. PMID: 27814522; PMCID: PMC5111816.