How to train a CNN to 99% accuracy on MNIST in less than a second on a laptop

Overview

Training a NN to 99% accuracy on MNIST in 0.76 seconds

A quick study on how fast you can reach 99% accuracy on MNIST with a single laptop. Our answer is 0.76 seconds, reaching 99% accuracy in just one epoch of training. This is more than 200 times faster than the default training code from Pytorch. To see the final results, check 8_Final_00s76.ipynb. If you're interested in the process read on below for a step by step description of changes made.

The repo is organized into jupyter notebooks, showing a chronological order of changes required to go from initial Pytorch tutorial that trains for 3 minutes to less than a second of training time on a laptop with GeForce GTX 1660 Ti GPU. I aimed for a coordinate ascent like procedure, changing only one thing at a time to make sure we understand what is the source of improvements each time, but sometimes I bunched up correlated or small changes.

Requirements

Python3.x and Pytorch 1.8 (most likely works with >= 1.3). For fast times you'll need Cuda and a compatible GPU.

0_Pytorch_initial_2m_52s.ipynb: Starting benchmark

First we need to benchmark starting performance. This can be found in the file 0_Pytorch_initial_2m_52s.ipynb. Note the code downloads the dataset if not already present so reporting second run time. Trains for 14 epochs each run, average accuracy of two runs is 99.185% on test set, and the mean runtime is 2min 52s ± 38.1ms.

1_Early_stopping_57s40.ipynb: Stop early

Since our goal is to reach only 99% accuracy, we don't need the full training time. Our first modification is to simply stop training after the epoch we hit 99% test accuracy. This is typically reached within 3-5 epochs with average final accuracy of 99.07%, cutting training time to around a third of the original at 57.4s ± 6.85s.

2_Smaller_NN_30s30.ipynb: Reduce network size

Next we employ the trick of reducing both network size and regularization to speed up convergence. This is done by adding a 2x2 max pool layer after the first conv layer, reducing parameters in our fully connected layers by more than 4x. To compensate we also remove one of the 2 dropout layers. This reduces number of epochs we need to converge to 2-3, and training time to 30.3s ± 5.28s.

3_Data_loading_07s31.ipynb: Optimize Data Loading!

This is probably the biggest and most surprising time save of this project. Just by better optimizing the data loading process we can save 75% of the entire training run time. It turns out that torch.utils.data.DataLoader is really inefficient for small datasets like MNIST, instead of reading it from the disk one batch at a time we can simply load the entire dataset into GPU memory at once and keep it there. To do this we save the entire dataset with the same processing we had before onto disk in a single pytorch array using data_loader.save_data(). This takes around 10s and is not counted in the training time as it has to be done only once. With this optimization, our average training time goes down to 7.31s ± 1.36s.

4_128_Batch_04s66.ipynb: Increase batch size

Now that we have optimized data loading, increasing batch size can significantly increase the speed of training. Simply increasing the batch size from 64 to 128 reduces our average train time to 4.66s ± 583ms.

5_Onecycle_lr_03s14.ipynb: Better learning rate schedule

For this step, we turn our looks to to the learning rate schedule. Previously we used an exponential decay where after each epoch lr is multiplied by 0.7. We replace this by Superconvergence also known as OneCycleLR, where the learning starts close to 0 and is linearly(or with cosine schedule) increased to to its peak value at the middle of training and slowly lowered down to zero again in the end. This allows using much higher learning rates than otherwise. We used peak LR of 4.0, 4 times higher than the starting lr used previously. The network reaches 99% in 2 epochs every time now, and this takes our training time down to 3.14s ± 4.72ms.

6_256_Batch_02s31.ipynb: Increase batch size, again

With our better lr schedule we can once more double our batch size without hurting performance much. Note this time around it doesn't reach 99% on all random seeds but I count it as a success as long I'm confident the mean accuracy is greater than 99%. This is because Superconvergence requires a fixed length training and we can't quarantee every seed works. This cuts our training time down to 2.31s ± 23.2ms.

7_Smaller_NN2_01s74.ipynb: Remove dropout and reduce size, again

Next we repeat our procedure from step 2 once again, remove the remaning dropout layer and compensate by reducing the width of our convolutional layers, first to 24 from 32 and second to 32 from 64. This reduces the time to train an epoch, and even nets us with increased accuracy, averaging around 99.1% after two epochs of training. This gives us mean time of 1.74s ± 18.3ms.

8_Final_00s76.ipynb: Tune everything

Now that we have a fast working model and we have grabbed most of the low hanging improvements, it is time to dive into final finetuning. To start off, we simply move our max pool operations before the ReLU activation, which doesn't change the network but saves us a bit of compute.

The next changes were the result of a large search operation, where I tried a number of different things, optimizing one hyperparameter at a time. For each change I trained on 30 different seeds and measured what gives us the highest mean accuracy. 30 seeds was necessary to make statistically significant conclusions on small changes, and it is worth noting training 30 seeds took less than a minute at this point. Higher accuracy can then be translated into faster times by cutting down on the number of epochs.

First I actually made the network bigger in select places that didn't slow down performance too much. The kernel size of first convolutional layer was incresed from 3 to 5, and the final fully connected layer increased from 128 to 256.

Next, it was time to change the optimizer. I found that with proper hyperparameters, Adam actually outperforms Adadelta which we had used so far. The hyperparameters I changed from default are learning rate of 0.01(default 0.001), beta1 of 0.7(default 0.9) and bata2 of 0.9(default 0.999).

All of this lead to a large boost in accuracy(99.245% accuracy after 2 epochs), which I was able to finally trade into faster training times by cutting training down to just one epoch! Our final result is 99.04% mean accuracy in just 762ms ± 24.9ms.

Owner
Tuomas Oikarinen
PhD student at UC San Diego, trying to understand ML and hopefully make it more safe. Previously @MIT.
Tuomas Oikarinen
[CVPR'22] Official PyTorch Implementation of Collaborative Transformers for Grounded Situation Recognition

[CVPR'22] Collaborative Transformers for Grounded Situation Recognition Paper | Model Checkpoint This is the official PyTorch implementation of Collab

Junhyeong Cho 29 Dec 10, 2022
Official implementation of paper "Query2Label: A Simple Transformer Way to Multi-Label Classification".

Introdunction This is the official implementation of the paper "Query2Label: A Simple Transformer Way to Multi-Label Classification". Abstract This pa

Shilong Liu 274 Dec 28, 2022
Official PyTorch implementation of the paper "Graph-based Generative Face Anonymisation with Pose Preservation" in ICIAP 2021

Contents AnonyGAN Installation Dataset Preparation Generating Images Using Pretrained Model Train and Test New Models Evaluation Acknowledgments Citat

Nicola Dall'Asen 10 May 24, 2022
Local Similarity Pattern and Cost Self-Reassembling for Deep Stereo Matching Networks

Local Similarity Pattern and Cost Self-Reassembling for Deep Stereo Matching Networks Contributions A novel pairwise feature LSP to extract structural

31 Dec 06, 2022
Predictive Maintenance LSTM

Predictive-Maintenance-LSTM - Predictive maintenance study for Complex case study, we've obtained failure causes by operational error and more deeply by design mistakes.

Amir M. Sadafi 1 Dec 31, 2021
CLNTM - Contrastive Learning for Neural Topic Model

Contrastive Learning for Neural Topic Model This repository contains the impleme

Thong Thanh Nguyen 25 Nov 24, 2022
Official implementation for paper Knowledge Bridging for Empathetic Dialogue Generation (AAAI 2021).

Knowledge Bridging for Empathetic Dialogue Generation This is the official implementation for paper Knowledge Bridging for Empathetic Dialogue Generat

Qintong Li 50 Dec 20, 2022
Code for "Intra-hour Photovoltaic Generation Forecasting based on Multi-source Data and Deep Learning Methods."

pv_predict_unet-lstm Code for "Intra-hour Photovoltaic Generation Forecasting based on Multi-source Data and Deep Learning Methods." IEEE Transactions

FolkScientistInDL 8 Oct 08, 2022
Understanding and Improving Encoder Layer Fusion in Sequence-to-Sequence Learning (ICLR 2021)

Understanding and Improving Encoder Layer Fusion in Sequence-to-Sequence Learning (ICLR 2021) Citation Please cite as: @inproceedings{liu2020understan

Sunbow Liu 22 Nov 25, 2022
なりすまし検出(anti-spoof-mn3)のWebカメラ向けデモ

FaceDetection-Anti-Spoof-Demo なりすまし検出(anti-spoof-mn3)のWebカメラ向けデモです。 モデルはPINTO_model_zoo/191_anti-spoof-mn3からONNX形式のモデルを使用しています。 Requirement mediapipe

KazuhitoTakahashi 8 Nov 18, 2022
Machine learning library for fast and efficient Gaussian mixture models

This repository contains code which implements the Stochastic Gaussian Mixture Model (S-GMM) for event-based datasets Dependencies CMake Premake4 Blaz

Omar Oubari 1 Dec 19, 2022
A note taker for NVDA. Allows the user to create, edit, view, manage and export notes to different formats.

Quick Notetaker add-on for NVDA The Quick Notetaker add-on is a wonderful tool which allows writing notes quickly and easily anytime and from any app

5 Dec 06, 2022
Learning Calibrated-Guidance for Object Detection in Aerial Images

Learning Calibrated-Guidance for Object Detection in Aerial Images arxiv We propose a simple yet effective Calibrated-Guidance (CG) scheme to enhance

51 Sep 22, 2022
Code accompanying paper: Meta-Learning to Improve Pre-Training

Meta-Learning to Improve Pre-Training This folder contains code to run experiments in the paper Meta-Learning to Improve Pre-Training, NeurIPS 2021. P

28 Dec 31, 2022
Pretty Tensor - Fluent Neural Networks in TensorFlow

Pretty Tensor provides a high level builder API for TensorFlow. It provides thin wrappers on Tensors so that you can easily build multi-layer neural networks.

Google 1.2k Dec 29, 2022
Final report with code for KAIST Course KSE 801.

Orthogonal collocation is a method for the numerical solution of partial differential equations

Chuanbo HUA 4 Apr 06, 2022
Official implementation of the paper "Steganographer Detection via a Similarity Accumulation Graph Convolutional Network"

SAGCN - Official PyTorch Implementation | Paper | Project Page This is the official implementation of the paper "Steganographer detection via a simila

ZHANG Zhi 1 Nov 26, 2021
Code for our paper A Transformer-Based Feature Segmentation and Region Alignment Method For UAV-View Geo-Localization,

FSRA This repository contains the dataset link and the code for our paper A Transformer-Based Feature Segmentation and Region Alignment Method For UAV

Dmmm 32 Dec 18, 2022
Video Frame Interpolation without Temporal Priors (a general method for blurry video interpolation)

Video Frame Interpolation without Temporal Priors (NeurIPS2020) [Paper] [video] How to run Prerequisites NVIDIA GPU + CUDA 9.0 + CuDNN 7.6.5 Pytorch 1

YoujianZhang 31 Sep 04, 2022
Official implementation for NIPS'17 paper: PredRNN: Recurrent Neural Networks for Predictive Learning Using Spatiotemporal LSTMs.

PredRNN: A Recurrent Neural Network for Spatiotemporal Predictive Learning The predictive learning of spatiotemporal sequences aims to generate future

THUML: Machine Learning Group @ THSS 243 Dec 26, 2022