Publications

Papers Presented

1. Applications of Deep Learning to Autonomous Vehicles - Vatsal Srivastava, Sumit Binnani - Fifth International Conference on Business Analytics and Intelligence, Indian Institute of Management (IIM), Bangalore

Abstract: Self-driving or autonomous vehicles have expanded dramatically over the last few years and are undoubtedly a popular application of the recent developments in Deep Learning. One of the most important tasks that any such vehicle is faced with is the accurate perception of its surroundings which is achieved through the use of on-board cameras. In order to successfully use the information generated from the cameras, an efficient system of computer vision algorithms must be used. ​ Deep Learning has shown tremendous success in this task and many of the state-of-the-art algorithms are now powered by Deep Learning​ . Convolutional Neural Networks (CNNs) help in the task of classification of the various objects in the image like traffic signs and traffic lights. They can also be used for high quality steering angle prediction. Fully Convolutional Networks (FCNs) help in the tasks of Semantic Segmentation like identifying the pixels that correspond to the road and those that do not. They also help in recognition and classification​ ​ of​ ​ the​ ​ various​ ​ objects​ ​ on​ ​ the​ ​ road​ ​ like​ ​ pedestrians,​ ​ other​ ​ vehicles,​ ​ etc.

In this paper, we look at the techniques in Deep Learning that are currently being used in the industry for the successful navigation of autonomous vehicles. We explore some of the feature selection and pre-processing techniques to make the model robust and to reduce the size of the data. We then present an Deep Learning based approach that can perform various tasks like steering angle prediction, object detection, semantic segmentation and lane detection. We have used a simulator provided by Udacity, that mimics the driving lane for a vehicle, for training and testing our model. Our model is able to steer the vehicle for multiple laps around the track, identify the lanes, identify the traffic signs and also generalize well to the dynamic road conditions. We are also working with Udacity to deploy and integrate our model on a real vehicle​ ​ and​ ​ validate​ ​ the​ ​ accuracy​ ​ of​ ​ our​ ​ system.

Keywords: ​ ​ Autonomous ​ ​ Vehicle ​ , ​ ​ Deep ​ ​ Learning, ​ ​ Convolutional ​ ​ Neural ​ ​ Networks, ​ ​ Fully Convolutional ​ ​ Networks

2. Feature​ ​Selection​ ​in​ ​Sparse​ ​Matrices - Vatsal Srivastava, Rahul Kumar - Fifth International Conference on Business Analytics and Intelligence, Indian Institute of Management (IIM), Bangalore

Abstract: Feature selection, as a preprocessing step to machine learning, is effective in reducing dimensionality, removing irrelevant data, increasing learning accuracy, and improving result comprehensibility. There are two main approaches for feature selection: wrapper methods, in which the features are selected using the supervised learning algorithm, and filter methods, in which​ ​ the​ ​ selection​ ​ of​ ​ features​ ​ is​ ​ independent​ ​ of​ ​ any​ ​ learning​ ​ algorithm. However, most of these techniques use feature scoring algorithms that make some basic assumptions about the distribution of the data like normality, balanced distribution of classes, non-sparsity or dense data-set, etc. The data generated in real world rarely follows such strict criteria. In some cases such as digital advertising, the generated data matrix is actually very sparse and follows no distinct distribution. For this reason, we have come up with a new approach towards feature selection for cases where the data-sets do not follow the above mentioned assumptions. Our methodology also presents an approach to solve the problem of skewness of data. ​ The efficiency and effectiveness of our methods is then demonstrated by comparison with other well known techniques of statistics like ANOVA, mutual information, KL divergence, Fisher score, Bayes’ error, Chi-square, etc. The data-set used for validation is a ​ real-world user-browsing history data-set used for ad-campaign targeting. It has very high dimensions and is highly sparse as well. Our approach reduces the number of features to a significant​ ​ degree​ ​ without​ ​ compromising​ ​ on​ ​ the​ ​ accuracy​ ​ of​ ​ the​ ​ final​ ​ predictions.

Keywords: ​ ​ Feature ​ ​ Selection, ​ ​ Sparse-Matrices, ​ ​ Filters, ​ ​ Wrappers.