Sumon Biswas @ ISU

Sumon Biswas

Ph.D. Student, Research Assistant

Laboratory of Software Design

Department of Computer Science

Iowa State University

Phone: +1 (515) 708-6166

Email:

Directions: Google Map

Research Projects

Boa Boa is a framework and domain-specific language for mining millions of software repositories and analyze programs in large-scale. Mainly, I have worked on adding the Python language support for Boa. We have leveraged Eclipse DLTK parser to build the Boa compiler for Python. Furthermore, I have worked on generating several Boa datasets by mining open source software from GitHub, especially the machine learning software projects.
D4 I am actively working in the D4 (Dependable Data Driven Discovery) Institute, which is a TRIPODS project funded by NSF. Our research focuses on ensuring dependability of data science software using programming language methodologies and software engineering techniques. Especially, my research intends to improve the fairness of machine learning models. Recently, I have published a paper in ESEC/FSE, where I have conducted an empirical study to investigate societal bias present in real-world ML models, and different mitigation algorithms.
SE for AI/ML Artificial intelligence (AI) and machine learning (ML) models are increasingly being used in software systems. However, we still lack software engineering (SE) support for the development lifecycle of such AI/ML based systems. For example, how can we effectively detect and repair bugs in ML models? Currently, our research is focusing on the data science pipeline and eying to propose SE methods for enhancing ML development in practice.
Empirical SE Availability of large number of open source software in GitHub, SourceForge and discussions in Q&A sites like StackOverflow, allows large-scale analysis and generate insights for different SE tasks such as testing, bug repair, code review, code comprehension, etc. We are mining a number of interesting patterns from bug-fixing commits, code comments, naming conventions, issues, pull requests, to develop methods and automated tools for the aid of software development.

Academic & Skill Based Software Projects

S: Supervised I: Worked individually G: Worked in a Group

S Supervised Software Projects as TA Selected Projects: NextGen 911, Game X, CyChat, Battle of the Worlds, CyDisk, PrePark, Songusoid, Smart Art, CyBike, Buisiness QR, Twenty One
G Uncertainty in DNN Hyperparameter Optimization In this project, we provided aid to the deep learning programmers to quantify uncertainty in the model hyperparameters and help them to make informed decision while initializing hyperparameters. We have leveraged a first order type Uncertain<T> to represent the uncertainty in the random hyperparameters and choose the best value by performing statistical tests on the distribution.
I New Semantics using Lambda Calculus I have extended Lambda calculus using Coq proof assistant to implement a core language that includes two Python language features (compound comparison statement and generator function). I have also described the syntax, operational semantics and type system of the features.
I Near Duplicate Detection Using Simhash Given millions of documents, for each file, rank the similar documents. I have preprocessed the documents, created weighted word vectors and then implemented Simhash (locality sensitive hashing algorithm to evaluate approximate cosine similarity) to generate 64-bit fingerprint of each document. Finally block permuted hamming search was implemented in the fingerprint space to find the near duplicate.
I Naive Bayes Classifier for Text Documents Classified text documents from a large pool using machine learning techniques. A Naive Bayes classifier was built to classify approximately 20,000 newsgroup documents. The original dataset is available here.
G Performance Benchmarking for Link Prediction Algorithms in Social Networks From a given snapshot of a social network database, we can predict whether a person can be potentially connected to another person, by analyzing existing links. We take two datasets (Facebook dataset from Stanford Large Network Dataset Collection and bibliography dataset from DBLP) and import that into MySQL, and Neo4J (Graph based DB) to evaluate the metrics for different network topology.
I Public Key Encryption (PEKS) with Bloom Filter PEKS with Keyword Search is one of the most used method to search keywords over encrypted data. However, semantic security is not preserved and dictionary attack can help attackers to guess keywords. This project solves that issue implementing PEKS and searching keywords over Bloom Filter.
I Path ORAM (Oblivious Random Access Memory) While using cloud platform or any other insecure memory, attack can be made using the access pattern. Oblivious RAM is the way to hide the memory access pattern with some extra bandwidth and memory overhead. Path ORAM uses a binary tree to store all memory blocks and changes the location of the blocks repeatedly to access the whole branch for a single block.
G Tank Battle (Android Shooting Game) While using cloud platform or any other insecure memory, attack can be made using the access pattern. Oblivious RAM is the way to hide the memory access pattern with some extra bandwidth and memory overhead. Path ORAM uses a binary tree to store all memory blocks and changes the location of the blocks repeatedly to access the whole branch for a single block.
G Doctors Window (A solution for Doctors and Patients) This is a complete solution for private practitioner doctors. The doctors can manage patients appointment automatically, generate electronic prescriptions and store patients database. The patient side app helps to schedule appointment, get notification and view medical history.