ajayn1997

3D-Shape-Generation-using-3D-GANS

An implementation of the paper "Learning a Probabilistic Latent Space of Object Shapes via 3D Generative-Adversial Modelling" by Wu (et al) presented at NIPS 2016. The paper introduces 3D GANs, which leverages volumetric convolutional networks and vanilla GANs to produce 3D objects from a probabilistic space. This implementations uses python and the Keras framework to build the 3D GAN Architecture.

Automated-Image-Captioning-with-Visual-Attention

An python implementation of a Automated Image Captioning with Visual Attention. The Image Captioning is done through an encoder-decoder architecture where the encoder is a deep CNN and the decoder is a RNN unit. A time based visual attention as described in neural machine translation systems is used by the RNN to produce very realistic captions for the images. This work is inspired by the paper " Show, Attend and Tell: Neural Image CaptionGeneration with Visual Attention" by Xu et. al.

ChatBot_seq2seq

A python based chat-bot based on deep seq2seq model trained to talk and interact like a friend. The system uses a encoder-decoder architecture with each block being a LSTM model. The models were trained on the Movie Dialog dataset and the end product was an interactive python app which could hold a good conversation with a human.

Face-Recognition-using-Siamese-Networks

This project aims to perform reliable facial recognition using minimal training data. An example of few-shot learning, this algorithm is able to distinguish between faces by using a dataset having only 10 instances of each face. This project uses Siamese networks to compute the similarity between 2 facial images and classify them accordingly. The algorithm implemented can be easily extended to many real world use cases like signature verification, text similarity with only minimal data points.

Face_Recognition

Python program to recognise faces and smiles

Music-generation-using-deep-learning

This project aims to create realistic music in abc-notation using char-RNN architecture. The char-RNN is trained on a corpus of musical notes encoded using the abc-notation. The trained network is able to generate new musical notes in the abc-notation. This project uses the concepts of generating text using RNNs and applies it to create a textual representation of music.

Neural-Architecture-Search-using-Reinforcement-Learning

An implementation of neural architecture search using the REINFORCE algorithm. we use a re-current network to generate the model descriptions of neural networks and trainthis RNN with reinforcement learning to maximize the expected accuracy of thegenerated architectures on a validation set. This algorithm is tested on the CIFAR-10 dataset. The project is inspired from the work presented in the paper "NEURAL ARCHITECTURE SEARCH WITH REINFORCEMENT LEARNING" by Barret et al from Google Brain.

Predicting-Future-Stock-Prices-using-Actor-Critic-Method

An implementation of a stock trading bot using an Actor Critic algorithm. The trading environment is converted into an MDP with the state being the stock values and actions of the agent is to either HOLD, SELL or BUY. The agent is trained to maximise the overall revenue in the simulated Trading environment.

RL-VRP-PtrNtwrk

Reinforcement Learning for Solving the Vehicle Routing Problem

Self-Critical-Sequential-training-with-RL-for-chatbots

A chatbot implemented as a seq-2-seq model and trained using cross entropy method. The performance of the chatbot is improved by using Sequence Level Training using REINFORCE algorithm. In order to apply the REINFORCE algorithm (Williams, 1992; Zaremba & Sutskever, 2015) to the problem of sequence generation we cast our problem in the reinforcement learning (RL) framework (Sutton & Barto, 1988). Our generative model (the RNN) can be viewed as an agent, which interacts with the external environment (the words and the context vector it sees as input at every time step). The parameters of this agent defines a policy, whose execution results in the agent picking an action. In the sequence generation setting, an action refers to predicting the next word in the sequence at each time step. After taking an action the agent updates its internal state (the hidden units of RNN). Once the agent has reached the end of a sequence, it observes a reward. We can choose any reward function. Here, we use BLEU (Papineni et al., 2002) and ROUGE-2 (Lin & Hovy, 2003) since these are the metrics we use at test time.

ajayn1997/RL-VRP-PtrNtwrk

Reinforcement Learning for Solving the Vehicle Routing Problem

ajayn1997/Neural-Architecture-Search-using-Reinforcement-Learning

An implementation of neural architecture search using the REINFORCE algorithm. we use a re-current network to generate the model descriptions of neural networks and trainthis RNN with reinforcement learning to maximize the expected accuracy of thegenerated architectures on a validation set. This algorithm is tested on the CIFAR-10 dataset. The project is inspired from the work presented in the paper "NEURAL ARCHITECTURE SEARCH WITH REINFORCEMENT LEARNING" by Barret et al from Google Brain.

ajayn1997/3D-Shape-Generation-using-3D-GANS

An implementation of the paper "Learning a Probabilistic Latent Space of Object Shapes via 3D Generative-Adversial Modelling" by Wu (et al) presented at NIPS 2016. The paper introduces 3D GANs, which leverages volumetric convolutional networks and vanilla GANs to produce 3D objects from a probabilistic space. This implementations uses python and the Keras framework to build the 3D GAN Architecture.

ajayn1997/ChatBot_seq2seq

A python based chat-bot based on deep seq2seq model trained to talk and interact like a friend. The system uses a encoder-decoder architecture with each block being a LSTM model. The models were trained on the Movie Dialog dataset and the end product was an interactive python app which could hold a good conversation with a human.

ajayn1997/Self-Critical-Sequential-training-with-RL-for-chatbots

A chatbot implemented as a seq-2-seq model and trained using cross entropy method. The performance of the chatbot is improved by using Sequence Level Training using REINFORCE algorithm. In order to apply the REINFORCE algorithm (Williams, 1992; Zaremba & Sutskever, 2015) to the problem of sequence generation we cast our problem in the reinforcement learning (RL) framework (Sutton & Barto, 1988). Our generative model (the RNN) can be viewed as an agent, which interacts with the external environment (the words and the context vector it sees as input at every time step). The parameters of this agent defines a policy, whose execution results in the agent picking an action. In the sequence generation setting, an action refers to predicting the next word in the sequence at each time step. After taking an action the agent updates its internal state (the hidden units of RNN). Once the agent has reached the end of a sequence, it observes a reward. We can choose any reward function. Here, we use BLEU (Papineni et al., 2002) and ROUGE-2 (Lin & Hovy, 2003) since these are the metrics we use at test time.

ajayn1997/Face_Recognition

Python program to recognise faces and smiles

ajayn1997/Oauth2.0Spring

ajayn1997/Predicting-Future-Stock-Prices-using-Actor-Critic-Method

An implementation of a stock trading bot using an Actor Critic algorithm. The trading environment is converted into an MDP with the state being the stock values and actions of the agent is to either HOLD, SELL or BUY. The agent is trained to maximise the overall revenue in the simulated Trading environment.

ajayn1997/Sequence-to-sequence-Video-Captioning-System

An implementation of a sequence-to-sequence video captioning system inspired by the paper "Sequence to Sequence – Video to Text" by Subhashini Et. Al. An end-to-end sequence-to-sequence model is used to generate captions for videos. For this we exploit recurrent neural networks, specifically LSTMs, which have demonstrated state-of-the-art performance in image caption generation. The LSTM model is trained on video-sentence pairs and learns to associate a sequence of video frames to a sequence of words in order to generate a description of the event in the video clip.

ajayn1997/Visual-Question-Answering-System

An end-to-end VQA system implemented using the Keras framework.Visual Question Answering (VQA) is one such challenge which requires high-level scene interpretation from images combined with language modelling of relevant Q&A. Given an image and a natural language question about the image, the task is to provide an accurate natural language answer. This is an implementation of the model proposed by the original VQA paper by Agrawal et. al.

ajayn1997/Automated-Image-Captioning-with-Visual-Attention

An python implementation of a Automated Image Captioning with Visual Attention. The Image Captioning is done through an encoder-decoder architecture where the encoder is a deep CNN and the decoder is a RNN unit. A time based visual attention as described in neural machine translation systems is used by the RNN to produce very realistic captions for the images. This work is inspired by the paper " Show, Attend and Tell: Neural Image CaptionGeneration with Visual Attention" by Xu et. al.

ajayn1997/Face-Recognition-using-Siamese-Networks

This project aims to perform reliable facial recognition using minimal training data. An example of few-shot learning, this algorithm is able to distinguish between faces by using a dataset having only 10 instances of each face. This project uses Siamese networks to compute the similarity between 2 facial images and classify them accordingly. The algorithm implemented can be easily extended to many real world use cases like signature verification, text similarity with only minimal data points.

ajayn1997/Music-generation-using-deep-learning

This project aims to create realistic music in abc-notation using char-RNN architecture. The char-RNN is trained on a corpus of musical notes encoded using the abc-notation. The trained network is able to generate new musical notes in the abc-notation. This project uses the concepts of generating text using RNNs and applies it to create a textual representation of music.

ajayn1997/BankChurnPredictor

ajayn1997/CNN_Numpy

A python based implementation of the CNN algorithm using Numpy.

ajayn1997/Face-Aging-with-conditional-GANs

An implementation of the paper title "Face Aging With Conditional Generative Adversarial Networks" by Grigory Antipov et al. An conditional GAN architecture is used in this project for automatic face aging. Significant steps include latent vector approximation with a Encoder for image recontruction. The novelty in the project is the "Identity Preserving" optimization of the latent space so that the latent vector retains the identity of the original image. This latent vector along with a conditioning vector which specifies the target age is used by the GAN to produce realistic projections of future or past facial images of an individual.

ajayn1997/Google-Stock-Price-Predictor

A real time financial model using deep RNN

ajayn1997/Reasearch-Proposals

Repository containing all detailed documents of research ideas/ problems

ajayn1997/SSD_Object_Detection

A python based implementation of the SSD algorithm for object detection from videos.