Karthik Yearning Deep Learning

Summary - Gradient Descent Finds Global Minimum

Gradient Descent Finds Global Minima of Deep Neural Networks Simon S. Du Jason D. Lee Haochuan Li Liwei Wang Xiyu Zhai Abstract Gradient descent finds a global minimum in training deep neural networks despite the objective function being non-convex. The current paper proves gradient descent achieves zero training loss in polyn... Read more

Summary - Mask RCNN

MASK R-CNN Facebook AI Research (FAIR) Kaiming He Georgia Gkioxari Piotr Dollar Ross Girshick What does Mask R-CNN do We present a conceptually simple, flexible, and general framework for object instance segmentation. Our approach efficiently detects objects in an image while simultaneously generating a high-quality segmentation... Read more

Interesting projects at Microsoft Research AI

Project Frigatebird. Link Github Code Taking instantaneous decisions during an uncertain situation is near impossible for machines even when considering multi-level observations and their striving ability to learn complex policies. This progress is facilitated by the availability of abundant data, simulators such as games. These projects ... Read more

Summary - Google BERT

BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding Jacob Devlin Ming-Wei Chang Kenton Lee Kristina Toutanova Abstract What’s BERT BERT stands for Bidirectional Encoder Representations from Transformers. How BERT is unique BERT is designed to pre-train deep bidirectional representations by jointl... Read more

Deep Learning interview questions.

Difference between Deep Networks vs Shallow Networks ? Deep Networks - The number of hidden layers are more in deep networks with large number of parameters. Since there are higher number of parameters, higher the degree of non-linearity in the network. Hence this increases the capability to extract high level features. Shallow Networks - The ... Read more

Summary - Super Resolution

Fast and Accurate Image Super Resolution by Deep CNN with Skip Connection and Network in Network Current trend We propose a highly efficient and faster Single Image Super-Resolution (SISR) model with Deep Convolutional neural networks (Deep CNN). Deep CNN have recently shown that they have a significant reconstruction performance on single-... Read more

5 Intresting papers from Google AI - Nov01

One Model To Learn Them All   This paper demonstrates a single model to solve problems spanning from multiple domains. This model is trained on ImageNet , COCO dataset , a speech recognition corpus, and an English parsing task.   Fluid Annotation   A tool for image annotation. This is a model which perfor... Read more

Summary - Densenet

Densenet-Densely Connected Convolutional Networks Paper Let’s understand Densenet. In this paper, we embrace this observation and introduce the Dense Convolutional Network (DenseNet), which connects each layer to every other layer in a feed-forward fashion. Whereas traditional convolutional networks with L layers have L connections—one betw... Read more

Different Convolutions

Pointwise Convolution or 1x1 convolution It is a 1x1 convolution kernel. This helps in acquiring pixel by pixel level feature extraction. But the main purpose is, When the number channels in the previous layers needs to be shrinked , then we use 1x1 convolution. This reduces the problem of computational cost. For all the features lear... Read more