NALSM: Neuron-Astrocyte Liquid State Machine

Related tags

Deep LearningNALSM
Overview

NALSM: Neuron-Astrocyte Liquid State Machine

This package is a Tensorflow implementation of the Neuron-Astrocyte Liquid State Machine (NALSM) that introduces astrocyte-modulated STDP to the Liquid State Machine learning framework for improved accuracy performance and minimal tuning.

The paper has been accepted at NeurIPS 2021, available here.

Citation

Vladimir A. Ivanov and Konstantinos P. Michmizos. "Increasing Liquid State Machine Performance with Edge-of-Chaos Dynamics Organized by Astrocyte-modulated Plasticity." 35th Conference on Neural Information Processing Systems (NeurIPS 2021).

@inproceedings{ivanov_2021,
author = {Ivanov, Vladimir A. and Michmizos, Konstantinos P.},
title = {Increasing Liquid State Machine Performance with Edge-of-Chaos Dynamics Organized by Astrocyte-modulated Plasticity},
year = {2021},
pages={1--10},
booktitle = {35th Conference on Neural Information Processing Systems (NeurIPS 2021)}
}

Software Installation

  • Python 3.6.9
  • Tensorflow 2.1 (with CUDA 11.2 using tensorflow.compat.v1)
  • Numpy
  • Multiprocessing

Usage

This code performs the following functions:

  1. Generate the 3D network
  2. Train NALSM
  3. Evaluate trained model accuracy
  4. Evaluate trained model branching factor
  5. Evaluate model kernel quality

Instructions for obtaining/setting up datasets can be accessed here.

Overview of all files can be accessed here.

1. Generate 3D Network

To generate the 3D network, enter the following command:

python generate_spatial_network.py

This will prompt for following inputs:

  • WHICH_DATASET_TO_GENERATE_NETWORK_FOR? [TYPE M FOR MNIST/ N FOR NMNIST] : enter M to make a network with an input layer sized for MNIST/Fashion-MNIST or N for N-MNIST.
  • NETWORK_NUMBER_TO_CREATE? [int] : enter an integer to label the network.
  • SIZE_OF_LIQUID_DIMENSION_1? [int] : enter an integer representing the number of neurons to be in dimension 1 of liquid.
  • SIZE_OF_LIQUID_DIMENSION_2? [int] : enter an integer representing the number of neurons to be in dimension 2 of liquid.
  • SIZE_OF_LIQUID_DIMENSION_3? [int] : enter an integer representing the number of neurons to be in dimension 3 of liquid.

The run file will generate the network and associated log file containing data about the liquid (i.e. connection densities) in sub-directory

/ /networks/ .

2. Train NALSM

2.1 MNIST

To train NALSM model on MNIST, enter the following command:

python NALSM_RUN_MAIN_SIM_MNIST.py

This will prompt for the following inputs:

  • GPU? : enter an integer specifying the gpu to use for training.
  • VERSION? [int] : enter an integer to label the training simulation.
  • NET_NUM_VAR? [int] : enter the number of the network created in Section 1.
  • BATCH_SIZE? [int] : specify the number of samples to train at same time (batch), for liquids with 1000 neurons, batch size of 250 will work on a 12gb gpu. For larger liquids(8000), smaller batch sizes of 50 should work.
  • BATCHS_PER_BLOCK? [int] : specify number of batchs to keep in memory for training output layer, we found 2500 samples works well in terms of speed and memory (so for batch size of 250, this should be set to 10 (10 x 250 = 2500), for batch size 50 set this to 50 (50 x 50 = 2500).
  • ASTRO_W_SCALING? [float] : specify the astrocyte weight detailed in equation 7 of paper. We used 0.015 for all 1000 neuron liquids, and 0.0075 for 8000 neuron liquids. Generally accuracy peaks with a value around 0.01 (See Appendix).

This will generate all output in sub-directory

/ /train_data/ver_XX/ where XX is VERSION number.

2.2 N-MNIST

To train NALSM model on N-MNIST, enter the following command:

python NALSM_RUN_MAIN_SIM_N_MNIST.py

All input prompts and output are the same as described above for run file NALSM_RUN_MAIN_SIM_MNIST.py.

2.3 Fashion-MNIST

To train NALSM model on Fashion-MNIST, enter the following command:

python NALSM_RUN_MAIN_SIM_F_MNIST.py

All input prompts and output are the same as described above for run file NALSM_RUN_MAIN_SIM_MNIST.py.

Instructions for training other benchmarked LSM models can be accessed here.

3. Evaluate Trained Model Accuracy

To get accuracy of a trained model, enter the following command:

python get_test_accuracy.py

The run file will prompt for following inputs:

  • VERSION? [int] : enter the version number of the trained model

This will find the epoch with maximum validation accuracy and return the test accuracy for that epoch.

4. Evaluate Model Branching Factor

To compute the branching factor of a trained model, enter the following command:

python compute_branching_factor.py

The run file will prompt for following inputs:

  • VERSION? [int] : enter the version number of the trained model.

The trained model directory must have atleast one .spikes file, which contains millisecond spike data of each neuron for 20 arbitrarily selected input samples in a batch. The run file will generate a .bf file with same name as the .spikes file.

To read the generated .bf file, enter the following command:

python get_branching_factor.py

The run file will prompt for following inputs:

  • VERSION? [int] : enter the version number of the trained model.

The run file will print the average branching factor over the 20 samples.

5. Evaluate Model Kernel Quality

Model liquid kernel quality was calculated from the linear speration (SP) and generalization (AP) metrics for MNIST and N-MNIST datasets. To compute SP and AP metrics, first noisy spike counts must be generated for the AP metric, as follows.

To generate noisy spike counts for NALSM model on MNIST, enter the following command:

python NALSM_RUN_MAIN_SIM_MNIST_NOISE.py

The run file requires a W_INI.wdata file (the initialized weights), which should have been generated during model training.

The run file will prompt for the following inputs:

  • GPU? : enter an integer to select the gpu for the training simulation.
  • VERSION? [int] : enter the version number of the trained model.
  • NET_NUM_VAR? [int] : enter the network number of the trained model.
  • BATCH_SIZE? [int] : use the same value used for training the model.
  • BATCHS_PER_BLOCK? [int] : use the same value used for training the model.

The run file will generate all output in sub-directory

/ /train_data/ver_XX/ where XX is VERSION number.

To generate noisy spike counts for NALSM model on N-MNIST, enter the following command:

python NALSM_RUN_MAIN_SIM_N_MNIST_NOISE.py

As above, the run file requires 'W_INI.wdata' file. All input prompts and output are the same as described above for run file NALSM_RUN_MAIN_SIM_MNIST_NOISE.py.

After generating the noisy spike counts, to compute the SP and AP metrics for each trained model enter the following command:

python compute_SP_AP_kernel_quality_measures.py

The run file will prompt for inputs:

  • VERSION? [int] : enter the version number of the trained model.
  • DATASET_MODEL_WAS_TRAINED_ON? [TYPE M FOR MNIST/ N FOR NMNIST] : enter dataset the model was trained on. The run file will print out the SP and AP metrics.

Instructions for evaluating kernel quality for other benchmarked LSM models can be accessed here.
Owner
Computational Brain Lab
Computational Brain Lab @ Rutgers University
Computational Brain Lab
GitHub Repository
A curated list of long-tailed recognition resources.

Awesome Long-tailed Recognition A curated list of long-tailed recognition and related resources. Please feel free to pull requests or open an issue to

Zhiwei ZHANG 542 Jan 01, 2023
Fluency ENhanced Sentence-bert Evaluation (FENSE), metric for audio caption evaluation. And Benchmark dataset AudioCaps-Eval, Clotho-Eval.

FENSE The metric, Fluency ENhanced Sentence-bert Evaluation (FENSE), for audio caption evaluation, proposed in the paper "Can Audio Captions Be Evalua

Zhiling Zhang 13 Dec 23, 2022
PGPortfolio: Policy Gradient Portfolio, the source code of "A Deep Reinforcement Learning Framework for the Financial Portfolio Management Problem"(https://arxiv.org/pdf/1706.10059.pdf).

This is the original implementation of our paper, A Deep Reinforcement Learning Framework for the Financial Portfolio Management Problem (arXiv:1706.1

Zhengyao Jiang 1.5k Dec 29, 2022
MoCap-Solver: A Neural Solver for Optical Motion Capture Data

MoCap-Solver is a data-driven-based robust marker denoising method, which takes raw mocap markers as input and outputs corresponding clean markers and skeleton motions.

55 Dec 28, 2022
PySLM Python Library for Selective Laser Melting and Additive Manufacturing

PySLM Python Library for Selective Laser Melting and Additive Manufacturing PySLM is a Python library for supporting development of input files used i

Dr Luke Parry 35 Dec 27, 2022
IRON Kaggle project done while doing IRONHACK Bootcamp where we had to analyze and use a Machine Learning Project to predict future sales

IRON Kaggle project done while doing IRONHACK Bootcamp where we had to analyze and use a Machine Learning Project to predict future sales. In this case, we ended up using XGBoost because it was the o

1 Jan 04, 2022
Distributing Deep Learning Hyperparameter Tuning for 3D Medical Image Segmentation

DistMIS Distributing Deep Learning Hyperparameter Tuning for 3D Medical Image Segmentation. DistriMIS Distributing Deep Learning Hyperparameter Tuning

HiEST 2 Sep 09, 2022
Source code for "Progressive Transformers for End-to-End Sign Language Production" (ECCV 2020)

Progressive Transformers for End-to-End Sign Language Production Source code for "Progressive Transformers for End-to-End Sign Language Production" (B

58 Dec 21, 2022
An efficient PyTorch implementation of the winning entry of the 2017 VQA Challenge.

Bottom-Up and Top-Down Attention for Visual Question Answering An efficient PyTorch implementation of the winning entry of the 2017 VQA Challenge. The

Hengyuan Hu 731 Jan 03, 2023
[CVPR 2021] Counterfactual VQA: A Cause-Effect Look at Language Bias

Counterfactual VQA (CF-VQA) This repository is the Pytorch implementation of our paper "Counterfactual VQA: A Cause-Effect Look at Language Bias" in C

Yulei Niu 94 Dec 03, 2022
Covid-19 Test AI (Deep Learning - NNs) Software. Accuracy is the %96.5, loss is the 0.09 :)

Covid-19 Test AI (Deep Learning - NNs) Software I developed a segmentation algorithm to understand whether Covid-19 Test Photos are positive or negati

Emirhan BULUT 28 Dec 04, 2021
CDTrans: Cross-domain Transformer for Unsupervised Domain Adaptation

CDTrans: Cross-domain Transformer for Unsupervised Domain Adaptation [arxiv] This is the official repository for CDTrans: Cross-domain Transformer for

238 Dec 22, 2022
Phylogeny Partners

Phylogeny-Partners Two states models Instalation You may need to install the cython, networkx, numpy, scipy package: pip install cython, networkx, num

1 Sep 19, 2022
A visualisation tool for Deep Reinforcement Learning

DRLVIS - Visualising Deep Reinforcement Learning Created by Marios Sirtmatsis with the support of Alex Bäuerle. DRLVis is an application used for visu

Marios Sirtmatsis 1 Nov 04, 2021
Simple torch.nn.module implementation of Alias-Free-GAN style filter and resample

Alias-Free-Torch Simple torch module implementation of Alias-Free GAN. This repository including Alias-Free GAN style lowpass sinc filter @filter.py A

이준혁(Junhyeok Lee) 64 Dec 22, 2022
FCOS: Fully Convolutional One-Stage Object Detection (ICCV'19)

FCOS: Fully Convolutional One-Stage Object Detection This project hosts the code for implementing the FCOS algorithm for object detection, as presente

Tian Zhi 3.1k Jan 05, 2023
This is just a funny project that we want to see AutoEncoder (AE) can actually work to enhance the features we want

Funny_muscle_enhancer :) 1.Discription: This is just a funny project that we want to see AutoEncoder (AE) can actually work on the some features. We w

Jing-Yao Chen (Jacob) 8 Oct 01, 2022
A PyTorch implementation of "ANEMONE: Graph Anomaly Detection with Multi-Scale Contrastive Learning", CIKM-21

ANEMONE A PyTorch implementation of "ANEMONE: Graph Anomaly Detection with Multi-Scale Contrastive Learning", CIKM-21 Dependencies python==3.6.1 dgl==

Graph Analysis & Deep Learning Laboratory, GRAND 30 Dec 14, 2022
Official PyTorch Implementation of Hypercorrelation Squeeze for Few-Shot Segmentation, arXiv 2021

Hypercorrelation Squeeze for Few-Shot Segmentation This is the implementation of the paper "Hypercorrelation Squeeze for Few-Shot Segmentation" by Juh

Juhong Min 165 Dec 28, 2022
Code for our NeurIPS 2021 paper 'Exploiting the Intrinsic Neighborhood Structure for Source-free Domain Adaptation'

Exploiting the Intrinsic Neighborhood Structure for Source-free Domain Adaptation (NeurIPS 2021) Code for our NeurIPS 2021 paper 'Exploiting the Intri

Shiqi Yang 53 Dec 25, 2022