Non-linear motor control by local learning in spiking neural networks
Authors: Aditya Gilra, Wulfram Gerstner
ICML 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We show that the network learns an inverse model of the non-linear dynamics, i.e. it infers from state trajectory as input to the network, the continuous-time command that produced the trajectory. Connection weights are adjusted via a local plasticity rule that involves pre-synaptic firing and post-synaptic feedback of the error in the inferred command. We propose a network architecture, termed differential feedforward, and show that it gives a lower test error than other feedforward and recurrent architectures. We demonstrate the performance of the inverse model to control a two-link arm along a desired trajectory. |
| Researcher Affiliation | Academia | 1School of Computer and Communication Sciences, and Brain Mind Institute, School of Life Sciences, Ecole Polytechnique F ed erale de Lausanne, 1015 Lausanne EPFL, Switzerland. 2Now at: Institute for Genetics, University of Bonn, Kirschallee 1, 53115 Bonn, Germany. |
| Pseudocode | No | The paper describes the learning rules and network architecture in text and diagrams but does not provide any structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | Code for learning the inverse model using FOLLOW and employing it for closed-loop control, is available at https://github.com/adityagilra/FOLLOWControl. |
| Open Datasets | No | The paper states, 'random 2-dimensional torque input uγ(t), γ = 1, 2 is provided to the arm to generate random state trajectories x(t) analogous to motor babbling.' and 'Details and parameters for generating the random input, the arm dynamics and the fixed random network parameters are as in (Gilra & Gerstner, 2017)'. The data is generated rather than using a pre-existing publicly available dataset, and no access information is provided for the generated data. |
| Dataset Splits | No | The paper discusses 'training' and 'testing' but does not explicitly mention a 'validation' dataset or its split percentages/counts. |
| Hardware Specification | No | The paper does not specify any hardware details such as GPU/CPU models, memory, or cloud computing instance types used for running the experiments. |
| Software Dependencies | No | The paper does not provide specific version numbers for any software dependencies (e.g., programming languages, libraries, or frameworks). |
| Experiment Setup | Yes | At the end of 10,000 s of FOLLOW learning on the feedforward weights, at learning rate 2 10 4, the error had plateaued... This u is consistent with the delay due to two synaptic filtering time constants of 20 ms each from the input state to the inferred command. Thus, for all remaining simulations including Figure 2, we used = 50 ms and u = 50 ms. We used 900 neurons when testing other parameters or architectures (Figs. 3, 4, 5), but 11,000 neurons for the final applications (Figs. 2, 7). |