Tiny ImageNet

Tiny ImageNet is a smaller-scale version of the ImageNet dataset, commonly used to benchmark the performance of deep learning models under data-scarcity conditions. Current research focuses on improving model accuracy and robustness on this dataset, exploring techniques like dataset distillation, data augmentation (including image generation and transformations), and novel training methods such as companion learning and sharpness-aware minimization. These efforts aim to bridge the performance gap between convolutional neural networks and vision transformers on limited data, impacting both the development of more efficient models and the advancement of continual and federated learning paradigms.

Papers

May 2, 2022

Zebra: Memory Bandwidth Reduction for CNN Accelerators With Zero Block Regularization of Activation Maps
Hsu-Tung Shih, Tian-Sheuan Chang
Pruning Method Tiny ImageNet Major Challenge Bottleneck Activation Map CNN Accelerator

February 10, 2022

Feature-level augmentation to improve robustness of deep neural networks to affine transformations
Adrian Sandru, Mariana-Iuliana Georgescu, Radu Tudor Ionescu
Convolutional Neural Network Deep Neural Network Native Robustness Data Augmentation CNN Architecture Tiny ImageNet Feature Augmentation Image Transformation Left Corner Transformation

December 30, 2021

A Novel Generator with Auxiliary Branch for Improving GAN Performance
Seung Park, Yong-Goo Shin
Generative Adversarial Network Image Generation Tiny ImageNet Cramer Wold Generator

December 27, 2021

Vision Transformer for Small-Size Datasets
Seung Hoon Lee, Seunghyun Lee, Byung Cheol Song
Convolutional Neural Network Vision Transformer Data Set Tiny ImageNet Local Self Attention Semantic Patch

December 24, 2021

BMPQ: Bit-Gradient Sensitivity Driven Mixed-Precision Quantization of DNNs from Scratch
Souvik Kundu, Shikai Wang, Qirui Sun, Peter A. Beerel, Massoud Pedram
Deep Neural Network DNN Framework Scratch Project Mixed Precision Quantization Tiny ImageNet Gradient Coding Mixed Precision Training

Tiny ImageNet

Papers

Zebra: Memory Bandwidth Reduction for CNN Accelerators With Zero Block Regularization of Activation Maps

Feature-level augmentation to improve robustness of deep neural networks to affine transformations

A Novel Generator with Auxiliary Branch for Improving GAN Performance

Vision Transformer for Small-Size Datasets

BMPQ: Bit-Gradient Sensitivity Driven Mixed-Precision Quantization of DNNs from Scratch