Akshay Mittal

Final Year Undergraduate
Department of Computer Science and Engineering
Indian Institute of Technology, Kanpur
India

Email: akshaymittal [AT] live.com, amittal [AT] cse.iitk.ac.in, akshaymittal10 [AT] gmail.com
Phone: +91-9415-337-327

[ Research | Publications | Systems Projects | Robotics | Achievements | Courses Taken | Positions | CV ]

Education

* at the end of 8 Semesters

Interests

Data Mining, Machine Learning, Programming Languages

Research Experience

• PocketWeb: An Instant Web-Browser Prototype for Mobile Devices
  

  Mentored by Dr. Dimitrios Lymberopoulos, Sensing and Energy Research Group (SERG) at Microsoft Research, Redmond (May-Jul 2011)

  The high network latencies and limited battery life of mobile phones can make mobile web browsing a frustrating experience. Prior work has proposed trading memory capacity for lower web access latency and a more convenient data transfer schedule from an energy perspective by prefetching slowly-changing data (search queries and results) nightly, when the phone is charging. However, most web content is intrinsically much more dynamic and may be updated multiple times a day, thus eliminating the effectiveness of periodic updates.

  This paper addresses the challenge of prefetching dynamic web content in a timely fashion, giving the user an instant web browsing experience but without aggravating the battery lifetime issue. We start by analyzing the web access traces of 8,000 users, and observe that mobile web browsing exhibits a strong spatiotemporal signature, which is different for every user. We propose to use a machine learning approach based on stochastic gradient boosting techniques to efficiently model this signature on a per user basis. The machine learning model is capable of accurately predicting future web accesses and prefetching the content in a timely manner. Our experimental evaluation with 48,000 models trained on real user datasets shows that we can accurately prefetch 60% of the URLs for about 80-90% of the users within 2 minutes before the request. The system prototype (~5000 LOC) we built not only provides more than 80% lower web access time for more than 80% of the users, but it also achieves the same or lower radio energy dissipation by more than 50% for the majority of mobile users.

  [ abstract | pdf | talk ]

  [back to top]
• Mining Minimally Infrequent Itemsets using Pattern-Growth Paradigm and Residual Trees
  

  Mentored by Prof. Arnab Bhattacharya under the SURGE Programme at IIT Kanpur (May-Jul 2010)

  Itemset mining has been an active area of research due to its successful application in various data mining scenarios including finding association rules. Though most of the past work has been on finding frequent itemsets, infrequent itemset mining has demonstrated its utility in web mining, bioinformatics and other fields. In this paper, we propose a new algorithm based on the pattern-growth paradigm to find minimally infrequent itemsets. A minimally infre-quent itemset has no subset which is also infrequent. We also introduce the novel concept of residual trees. We fur-ther utilize the residual trees to mine multiple level mini-mum support itemsets where different thresholds are used for finding frequent itemsets for different lengths of the itemset. Finally, we analyze the behavior of our algorithm with respect to different parameters and show through ex-periments that it outperforms the competing ones.

  [ abstract | pdf | talk ]

  [back to top]
• Chemistry Studio: An Intelligent Tutoring System
  

  Bachelor of Technology Project under Dr. Sumit Gulwani (Microsoft Research Redmond), Dr. Ashish Tiwari (SRI International), Prof. Amey Karkare (IIT Kanpur) and Prof. Rupak Majumdar (University of California at Los Angeles - UCLA) (Aug 2011 - May 2012)

  In this paper, we present the prototype of an automated education system (~5500 LOC), aimed at helping the students in their learning process. Our main motivation is to introduce a dynamic component in such learning systems as compared to existing systems which are primarily static in nature. We aim to automatically solve problems, in the domain of Periodic Table and its properties, with proper logic and reasoning, generate solutions and explanations in accordance with the interest and knowledge of the student. The project involves formulation of a logical system for the Periodic Table in Prolog with an interface in C#. The system instead of being based on direct lookup paradigm, breaks a complex problem into certain basic facts by relating it to the underlying logic and provides a smooth flow of reasoning from the base argument up to the final solution. We extend this to an interactive system where a student can ask for preferential solution taking into account the knowledge base of a specific individual. We provide the feature of generating new problems which are similar to a given problem. The project's main features include logic formulation, problem solving in logic, hint generation and problem generation. We present a statistical evaluation showing that the system solves a large number of questions for each template that is created.

  [back to top]
• Extracting Topic Chains from News Articles
  

  Mentored by Prof. Arnab Bhattacharya, Dept. of CSE, IIT Kanpur (Sept-Nov 2011)

  The users around the globe are facing a crisis of information overload. The users find it difficult to keep up with the progress of various news stories over a period of time. The main cause is the overwhelming amount of news and updates posted daily by the media. With this much data, it is often easy to miss the big picture.

  In this paper, we address the issue of retrieving coherent chains between given two news articles. We devise a modified version of Dijkstra Algorithm for finding the shortest path between two document, keeping in mind the coherent nature of the chain and not just the pairwise coherency. We adapt a similarity measure based on the correlation of documents through words. We also test our algorithm with other standard similarity measure. We propose two promising approaches which can extended further: using Semantic similarity along-with Earth Movers and using Kullback Leibler divergence on topic distribution of the documents. We evaluate our techniques by soliciting users and taking feedback. We also devise an evaluation metric which captures the amount of incorrect articles given with respect to a manually generated link chain.

  [back to top]
• PageRank for Product Image Search
  Paper presentation in the course Data Mining under Prof. Arnab Bhattacharya, Dept. of CSE, IIT Kanpur (Oct 2011)
  • Presented the paper "PageRank for Product Image Search" by Jing et al.
  • Awarded one of the best paper presentations
  [ talk | pdf ]


• Recognizing Text from CAPTCHA
  

  Course Project in Machine Learning and Knowledge Discovery under Prof. Krithika Venkataramani, IIT Kanpur (Jan-Apr 2011)

  In this paper, we present application of machine learning techniques to the problem of reading text from CAPTCHAs. We focus on Eliot's CAPTCHA for this purpose. We first present techniques for reading a simple CAPTCHA that requires only single character classification and trivial segmentation. We then show how we use more complex seg-mentation approach along with clustering techniques to read more difficult CAPTCHA; one that requires significant effort for noise removal and cluttering before classification of individual characters can be attempted. We detail our noise removal and clustering-based techniques, and present our results for this CAPTCHA.

  [ pdf | talk | code ]

  [back to top]

Publications

Systems Projects

• Synthesis of Incremental Excel Macros
  Course Project in Program Analysis, Verification & Testing under Prof. Subhajit Roy, Dept. of CSE, IIT Kanpur and Dr. Sumit Gulwani (MSR) (Feb-Apr 2012)
  • Extracted macros from the collection of EUSES spreadsheets (~5600) (a C# script ~250 LOC)
  • Designed incremental programs for basic mathematical operations like max, sum, etc. w.r.t. different change models
  • Analyzed the utility of basic incremental macros in the generation of incremental programs for complex macros that involve composition of the above basic operations
  [ pdf | code ]


• Compiler for Object Pascal
  Course Project in Compiler Design under Prof. Amey Karkare, Dept. of CSE, IIT Kanpur (Jan-Apr 2011)
  • Supports arrays, records, pointers, nested/recursive/overloaded procedures/functions, return value, call by reference
  • Implemented the compiler in Python (~3700 LOC) and produced assembly code for MIPS 32-bit architecture as output
  [ pdf | talk | code ]


• Grid Navigation/Mapping Agent
  Course Project in Artificial Intelligence Programming under Prof. Harish Karnick, Dept. of CSE, IIT Kanpur (Jan-Apr 2011)
  • Coded an agent in Python (~1000 LOC) that navigated and mapped a dynamic arena of objects
  • The objects could move with small probability on interaction with the agent
  [ pdf | code ]


• Extension of NACHOS
  Course Project in Operating Systems under Prof. Sumit Ganguly, Dept. of CSE, IIT Kanpur (Aug-Nov 2010)
  • Extended an instructional operating system NACHOS (Java) to implement various operating system concepts
  • Implemented and evaluated algorithms for Thread Scheduling, Page Replacement, Synchronization


• Building a Firewall
  Course Project in Computer Networks under Prof. Dheeraj Sanghi, Dept. of CSE, IIT Kanpur (Sept-Nov 2010)
  • Developed a firewall (~1300 LOC) that blocked unauthorized access while permitting authorized communications
  • Configured its devices to permit, deny, or proxy all computer traffic between different security domains
  • Used packet filtering for analysis of the data packets received and the implementation language being C
  • Transferred data packets back-and-forth successfully between authorized machines
  [ pdf | code ]


• New Set Operations for the MySQL Database Server (Minus, Intersect, Divide)
  Course Project in Principles of Data Base Systems under Prof. Arnab Bhattacharya, IIT Kanpur (Jan-Apr 2010)
  • Designed and developed a PHP wrapper (~800 LOC) for performing the set operations - minus[all], intersect[all], divide[all]
  • MySQL server lacks non-trivial implementations of these set operations, hence the motivation for the project
  • Parsed the input query having these set operations and generated corresponding complex queries for MySQL
  • Supports nested queries and multi-set operations similar to the set operation union, besides all the normal queries
  [ pdf | talk | code ]

  [back to top]

Robotics

• Construction Robot
  Done under the Robotics Club at IIT Kanpur (Aug-Sept 2009)
  • Built an autonomous robot picked blocks of thermocol and placed them at desired places to build a pi-shaped model
  • Incorporated an effective grabbing and placing mechanism along with infra-red digital and analog sensors
  • Won First prize in the event Construction Robot in Takneek '09 (annual inter-hostel technical festival of IIT Kanpur)


• Intercommunicating Typewriter Robots
  Done under the Robotics Club at IIT Kanpur (May-Jun 2009)
  • Built two autonomous robots for drawing 2D figures with independent control of the two degrees of freedom
  • Autonomous intercommunication and synchronization used for predefined figures, told only to the "Master Bot"
  • Used C as the implementation language for generating the hex files and intercommunication was achieved via UART
  [ pdf | pics ]


• Park Ur Porsche Robot
  Done under the Robotics Club at IIT Kanpur (Jan-Feb 2009)
  • Built an autonomous vehicle that parked itself in a parking lot amidst several cars using infra-red analog sensors
  • Won First prize in the event Park Ur Porsche in Techkriti '09 (the annual inter-college technical festival of IIT Kanpur)
  • Goal of the event was to park/un-park the robot autonomously in as many parking slots of the arena as possible
  [ video | pics ]


• Line Ballower Robot
  Done under the Robotics Club at IIT Kanpur (Aug-Sept 2008)
  • Built an autonomous robot that followed a white-line path using digital sensors, grabbed a ball & dropped in a basket
  • Used TSOP-1738 sensors to guide the locomotion and L298 & L293 motor drivers for the grabbing mechanism
  • Won First prize in the event Line Ballower in Takneek '08 (the annual inter-hostel technical festival of IIT Kanpur)
  [ pdf | video | pics ]

  [back to top]

Academic Achievements

Courses Taken

Positions of Responsibility

• Assistant Coordinator of Robogames in Techkriti 2010, the technical festival of IIT Kanpur
• Student Guide of the Counseling Service, IIT Kanpur for the year 2009-10
• Secretary of Robotics Club, IIT Kanpur for the year 2009-10
• Secretary, Software Corner in Techkriti 2009
• Held the post of School-Vice Captain and House Captain for the year 2007-08