The article describes a bit about how i got into research and how i started building TReX to help myself and all researchers out there.

July 2019, i got my offer letter from Microsoft Research to join them as a Research Intern. Couldn’t be more happier. Didn’t really know that i was going to work on a paper. Had little idea about how research works. I read research papers back then too. A lot of them. Didn’t understand many things but i kept reading them anyway. …

Summary of the paper
DeepFool: A Simple and Accurate Method to Fool Deep Neural Networks
by Seyed-Mohsen Moosavi-Dezfooli, Alhussein Fawzi, Pascal Frossard
Link to the paper: https://arxiv.org/pdf/1511.04599.pdf

Overview

Deep Neural Networks achieve state of the art performances in many tasks but fail miserably on slightly perturbed images, perturbed in a meaningful way (and not randomly).

The DeepFool paper have the following major contributions:

1. Simple and accurate method for computing the robustness of different classifiers to adversarial perturbations.
2. Experiments showing
   - DeepFool computes a more optimal adversarial perturbation
   - Adversarial Training significantly increases the robustness.

DeepFool for binary classifiers

A paper summary of the paper
PVANET: Deep but Lightweight Neural Networks for Real-time Object Detection
by Xinyu Zhou, Cong Yao, He Wen, Yuzhi Wang, Shuchang Zhou, Weiran He, and Jiajun Liang
Link to the paper: https://arxiv.org/pdf/1608.08021.pdf

Overview

This paper presents our lightweight feature extraction network architecture for object detection, named PVANET, which achieves real-time object detection performance without losing accuracy.

1. Computational Cost: 7.9GMAC for feature extraction with 1065x640input
2. Runtime Performance: 750ms/image (1.3FPS) on Intel i7 and 42ms/image (21.7FPS) on NVIDIA Titan X GPU
3. Accuracy: 83.8% mAP on VOC-2007; 82.5% mAP on VOC-2012

The key design principle is “less channels with…

This is a paper summary of the paper:
ENet: A Deep Neural Network Architecture for Real-Time Semantic Segmentation
by Adam Paszke
Paper: https://arxiv.org/abs/1606.02147

Overview

ENet (Efficient Neural Network) gives the ability to perform pixel-wise semantic segmentation in real-time. ENet is upto 18x faster, requires 75x less FLOPs, has 79x less parameters and provides similar or better accuracy to existing models. Tested on CamVid, CityScapes and SUN datasets.

This is a chapter summary from the one of the most popular Reinforcement Learning book by Richard S. Sutton and Andrew G. Barto (2nd Edition). The book can be found here: Link.

Reinforcement Learning is learning what to do — how to map situations to actions — so as to maximize a numerical reward signal. A learning agent can take actions that affect the state of the environment and have goals relating to the state of the environment. One of the challenges that arise in Reinforcement Learning, and not in other kinds of learning, is trade-off between exploration and exploitation…

This is a paper summary of the paper:
Deep Learning for Classical Japanese Literature
by Tarin Clanuwat, Mikel Bober-Irizar, Asanobu Kitamoto, Alex Lamb, Kazuaki Yamamoto, and David Ha.
Paper: https://arxiv.org/abs/1812.01718

Overview

The paper introduces 3 new benchmark datasets for Machine Learning, namely:
- Kuzushiji-MNIST — A drop-in replacement for MNIST dataset (28x28)
- Kuzushiji-49 — A much larger but imbalanced dataset containing 48 Hiragana characters and 1 Hiragana iteration mark (28x28)
- Kuzushiji-Kanji — An imbalanced dataset of 3832 Kanji characters, including rare characters with very few samples. (64x64)

In this article, we will see what are Convolutional Neural Networks, ConvNets in short. ConvNets are the superheroes that took working with images in deep learning to the next level. With ConvNets, the input is a image, or more specifically, a 3D Matrix.

Let’s start by looking at how a ConvNet looks!

This is a paper summary of the paper:
Perceptual Losses for Real-Time Style Transfer and Super-Resolution
by Justin Johnson, Alexandre Alahi, Li Fei-Fei.
Paper: https://arxiv.org/pdf/1603.08155.pdf

Overview

This paper proposes the use of perceptual loss functions for training feed-forward networks for image transformation tasks, instead of using per-pixel loss functions.

Per-pixel loss functions?
Comparing two images based on their individual pixel values.
So, if two images, that are perceptually the same, but different from each other based on even one pixel, then based on per-pixel loss functions they will be very different from each other.

Perceptual loss functions? Comparing two images based…