Course Description: This course provides a broad introduction to machine learning and statistical pattern recognition. Topics include: supervised learning (generative/discriminative learning, parametric/non-parametric learning, neural networks, support vector machines); unsupervised learning (clustering, dimensionality reduction, kernel methods); learning theory (bias/variance tradeoffs; VC theory; large margins); reinforcement learning and adaptive control. The course will also discuss recent applications of machine learning, such as to robotic control, data mining, autonomous navigation, bioinformatics, speech recognition, and text and web data processing.
Course Link

Project: Summarization of sports videos
Analyzing videos based on hand crafted features is a tedious task and is highly dependent on the type of data. These methods cannot be easily extended to account for other sensor modalities like audio, texts. The aim of the project is to be able to use both video and audio in efficiently and effectively recognizing events of interest in sports videos. Using unsupervised deep learning we learn the features from the data itself and try to identify a set of predefined classes in sports videos (Goal, Penalty, Foul, FreeKick, Corner). We built a baseline system using hand crafted vision features like HOG3D and compared it against the performance of an unsupervised learning method (2-3 hidden layered ISA).
Final report can be found here - PDF.

Course Description: The course covered recent research on the structure and analysis of such large social and information networks and on models and algorithms that abstract their basic properties. The course explored how to practically analyze large scale network data and how to reason about it through models for network structure and evolution. Topics include methods for link analysis and network community detection, diffusion and information propagation on the web, virus outbreak detection in networks, and connections with work in the social sciences and economics.
Course Link

Project: Modeling and Analysis of Real World Networks using Kronecker Graphs
Kronecker Graphs have been shown to elegantly model real world networks, while being mathematically tractable. In this project, we explored the property of self-similarity exhibited by real world networks like Twitter and popular real world datasets like Memetracker. The goal of the paper is to analyze fractal and self-similar properties of real world and Kronecker graphs by applying methods such as Box counting, Bifurcation ratio using Horton-Strahler index, Hurst exponent.
Final report can be found here - PDF.

Course Description: This course introduces the fundamental concepts and ideas in natural language processing (NLP), otherwise known as computational linguistics. It develops an in-depth understanding of both algorithms for processing linguistic information and the underlying computational properties of natural languages. We consider word-level, syntactic, and semantic processing from both a linguistic and an algorithmic perspective, aiming to get up to speed with current research in the area. The course focuses on modern quantitative techniques in NLP -- using large corpora, statistical models for acquisition, disambiguation, and parsing -- and the construction of representative systems.
Course Link

Component: Language models using Europarl corpus
Final report can be found here - PDF.

Component: Word level alignment systems (Machine Translation)
Final report can be found here - PDF.

Component: Named entity recognition and parsing systems
Final report can be found here - PDF.

Project: Multi-document extraction based Summarization
In this paper, we present three techniques for generating extraction based summaries including a novel graph based formulation to improve on the former methods. The first method uses a sentence importance score calculator based on various semantic features and a semantic similarity score to select sentences that would be most representative of the document. It uses stack-decoder algorithm as used as a tem- plate and builds on it to produce summaries that are closer to optimal. The second approach clusters sentences based on the above semantic similarity score and picks a repre- sentative from each cluster to be included in the generated summary. The third approach is a novel graph problem based formulation where summaries are generated based on the cliques found in the constructed graph. The graph is generated by building edges between sentences which talk about similar topics but are semantically not similar.
Final report can be found here - PDF.

Course Description: This course provides a broad introduction to various data mining techniques. The course will discuss data mining and machine learning algorithms for analyzing very large amounts of data. The emphasis will be on Map Reduce as a tool for creating parallel algorithms that can process very large amounts of data.
Course Link

Project: Topic Chaining and Phrase Linking
In this project, we implemented a technique to break down a collection of news articles into semantically co- herent threads. The chaining of articles is done based on the content and temporal aspects of the news articles. The problem of computing threads was solved by using a matching based algorithm on a relevance graph. We also tried two approaches in analyzing the resulting threads to get relations between the most common phrases: (a) Timestamp based clustering to get phrase group links and (b) Matching on the graph constructed using phrases to get links.
Final report can be found here - PDF.

Course Description: An introduction to the concepts and applications of computer vision. Topics include: cameras and projection models, low-level image processing methods such as filtering and edge detection; mid-level vision topics such as segmentation and clustering; shape reconstruction from stereo, as well as high-level vision tasks such as object recognition, scene recognition, face detection and human motion categorization.
Course Link

Project: Find Mii Challenge
We describe the approaches taken to solve the 4 challenge problems: (a) Find this Mii, (b) Find 2 look-alikes, (c) Find n odd Miis out, and (d) Find the fastest Mii in the “FindMii” game. The approaches varied depending on the problem at hand and in this report, we would justify the reasons for the selection of a family of features and also explain the improvements implemented. We used HOG features and Boosted decision trees to train a new classifier, SIFT to match faces, optical flow to detect the direction and amount of motion, Mean shift clustering to group similar points together etc to successfully navigate the tasks.
Final report can be found here - PDF.

Course Description: Basic and advanced techniques for text-based information systems: efficient text indexing; Boolean and vector space retrieval models; evaluation and interface issues; Web search including crawling, link-based algorithms, and Web metadata; text/Web clustering, classification; text mining.
Course Link

Project: Classification of messages
In this project, we implemented a family of supervised machine learning methods to classify Usenet newsgroup messages.
Final report can be found here - PDF.

Course Description: Problems in game playing, natural language processing, computer vision, robotics are challenging due to the inherent noise/uncertainty and computational complexity. This course provides the mathematical and algorithmic framework for tackling these sorts of problems. Topics include search, decision theory, graphical models, machine learning, and various applications.
Course Link

Project: Object Recognition and Tracking
In this project, we implemented an object recognition and tracking program that identifies and labels occurrences of 5 objects (mug, stapler, keyboard, clock, and scissors) in a given video clip.
Final report can be found here - PDF.