0/1 Deep Neural Networks via Block Coordinate Descent

2022-06-19Unverified0· sign in to hype

HUI ZHANG, Shenglong Zhou, Geoffrey Ye Li, Naihua Xiu

Unverified — Be the first to reproduce this paper.

Abstract

The step function is one of the simplest and most natural activation functions for deep neural networks (DNNs). As it counts 1 for positive variables and 0 for others, its intrinsic characteristics (e.g., discontinuity and no viable information of subgradients) impede its development for several decades. Even if there is an impressive body of work on designing DNNs with continuous activation functions that can be deemed as surrogates of the step function, it is still in the possession of some advantageous properties, such as complete robustness to outliers and being capable of attaining the best learning-theoretic guarantee of predictive accuracy. Hence, in this paper, we aim to train DNNs with the step function used as an activation function (dubbed as 0/1 DNNs). We first reformulate 0/1 DNNs as an unconstrained optimization problem and then solve it by a block coordinate descend (BCD) method. Moreover, we acquire closed-form solutions for sub-problems of BCD as well as its convergence properties. Furthermore, we also integrate _2,0-regularization into 0/1 DNN to accelerate the training process and compress the network scale. As a result, the proposed algorithm has a high performance on classifying MNIST and Fashion-MNIST datasets. As a result, the proposed algorithm has a desirable performance on classifying MNIST, FashionMNIST, Cifar10, and Cifar100 datasets.

Tasks

10-shot image generation 16k 2D Object Detection 3D dense captioning 3D Face Alignment 3D Facial Expression Recognition 3D Facial Landmark Localization 3D Hand Pose Estimation 3D Instance Segmentation 3D Lane Detection 3D Multi-Object Tracking 3D Place Recognition Abstractive Text Summarization Action Recognition Anomaly Detection Arithmetic Reasoning Articles Asthmatic Lung Sound Classification Audio Classification Change Detection Classification Click-Through Rate Prediction Code Generation Color Image Denoising Common Sense Reasoning Cross-Domain Few-Shot Object Detection Deblurring DeepFake Detection Denoising Depth Estimation Domain Generalization Drug Discovery EEG 4 classes Face Detection Face Recognition Fake Image Detection Fine-Grained Image Classification Fracture detection Fraud Detection Gloss-free Sign Language Translation Graph Classification Handwritten Mathmatical Expression Recognition Hateful Meme Classification Highlight Detection Image Captioning Image Classification Image Dehazing Image Generation Keyword Spotting Language Modelling License Plate Detection Long-range modeling Low-Light Image Enhancement Machine Translation Medical Image Segmentation Meme Classification Monocular Depth Estimation Multi-Label Classification Multimodal Emotion Recognition Multimodal Intent Recognition Multi-Object Tracking Music Source Separation NavSim Novel View Synthesis Object Detection Object Detection In Aerial Images Object Rearrangement Object Tracking Person Re-Identification Phone-level pronunciation scoring Pose Estimation Question Answering Railway Track Image Classification Real-Time Object Detection Rgb-T Tracking Robot Manipulation Robot Manipulation Generalization Robot Task Planning Semantic Segmentation Speech Enhancement Speech Recognition Style Transfer Table-to-Text Generation Temporal Relation Extraction Text to 3D Text-to-Image Generation Universal Domain Adaptation Unsupervised Domain Adaptation Video deraining Video Generation Video Question Answering Virtual Try-on Visual Object Tracking Weakly Supervised Action Localization Zero-Shot Video Question Answer

0/1 Deep Neural Networks via Block Coordinate Descent

Abstract

Tasks

Benchmark Results

Reproductions