CSE 2022 Fall Departmental Demo Day

(DRONES Lab - Department of Arts collaboration Project)

Bio-inspired Robotic Oysters

• Stephanie Rothenberg
• Karthik Dantu
• Ralfy Chettiar
• Chetan Palkar
• Sanket Dhande

“Tending Ostreidae: Serenades for Settling,” is an interdisciplinary project exploring how anthropogenic noise in New York City’s busy waterways impacts marine organisms, specifically oysters. Since oyster reefs provide habitat for hundreds of species, prevent erosion along the shorelines, and can filter about 30 to 50 gallons of water every day, oyster restoration in New York Harbor is a priority.

(PhD Poster Competition)

Detcting Deepfake Audio/UB Media Forensic Lab

• Chengzhe Sun
• Siwei Lyu

Deepfake technologies, which allow malicious actors to produce fake images, videos, and audio clips, are reaching an unprecedented convergence of quality, scalability, and ease of use. It will soon be possible to mass produce highly realistic synthetic content that may be generated and spread faster than fake media detectors can manage. The proliferation of these technologies poses clear threats to society and democracy (for example, consider the dangers of shared videos wherein politicians give fake speeches). It appears that the future of information channels which we rely on when forming our beliefs and opinions, is on the shaky ground unless detection technology can gain the upper hand. Synthetic audio detection is one key element in managing this threat.

Darvis: A modular, extensible framework for modern visual SLAM systems

• Sofiya Semenova
• Pranay Meshram

Visual SLAM is a long-standing research area with many significant advances over the years. New systems typically build on previous contributions, but this requires significant development overhead, a highly detailed understanding of previous system implementations, and is rife with programming pitfalls. To enable fast experimentation and reduce the need for researchers to re-invent the wheel, we propose a modular, extensible Visual SLAM framework.

Metagenomic Classification on Mobile Devices

• Andrew Mikalsen

Metagenomic classification is a form of DNA analysis that aims to classify organisms based on their DNA fragments present in a biological sample. The recent introduction of portable DNA sequencers has fostered great interest in performing metagenomic classification in mobile settings, motivated by applications such as rapid medical diagnosis, detection and tracking of infectious diseases, automated bioterrorism response systems, monitoring dynamic microbial communities, and more. However, the software to perform metagenomic classification remains incompatible with mobile workflows. Metagenomic classification is extremely resource intensive in terms of compute, memory, energy, network, and storage, and typical tools are designed to operate in large data centers, on supercomputers, and in the cloud, whose resources far outclass those available to mobile devices. Despite the advancements in portable DNA sequencing technology, the question of how to design software systems for metagenomics that are capable of mobile deployments remains wholly unanswered. In this project, we introduces Coriolis, a metagenomic classifier designed from the ground up to be efficient in mobile environments. Coriolis is based on novel programming abstractions and data structures that enable efficient management of scarce resources on mobile devices. These programming abstractions are packaged in the form of SMARTEn, a programming model and framework for expressing metagenomic classification algorithms. SMARTEn decomposes the metagenomic classification process into individual tasks, opening up opportunities to automatically manage and optimize their execution at runtime. To manage the massive reference database of known genomes required for metagnomic classification, we introduce the compact string B-tree (CSBT) as a practical data structure for managing unstructured text on disk. The CSBT is a space-efficient and generalized variant of (Ferragina and Grossi, 1999)'s string B-tree (SBT). At a high level, the CSBT can be thought of as a B+-tree of suffix offsets where each node's key set is represented as a suffix array augmented with information on the mismatched characters between adjacent suffixes. We prove that this information is sufficient to identify a query string's rank in a CSBT node while performing only a single disk access, and we give an I/O optimal algorithm for finding all occurrences of the longest matching prefix of a pattern in the text. The resulting representation is an order of magnitude smaller than the SBT, resulting in far better throughput and space usage. Furthermore, additional techniques enable the CSBT to index multiple texts while eliminating the need to store duplicate information, significantly improving performance for practical applications. Rigorous theoretical and experimental analysis demonstrates that Coriolis outperforms the state-of-the-art and enables real-time in the field metagenomics.

ret2ns

• Zheyuan Ma
• Xi Tan
• Ziming Zhao

The ARM Cortex-M microcontroller architecture is designed for embedded and Internet of things (IoT) applications. To facilitate efficient execution, it has some unique hardware optimization. For example, Cortex-M TrustZone has a fast state switch mechanism that allows direct control-flow transfer from secure privileged mode to non-secure unprivileged mode. In this paper, we demonstrate how this fast state switch mechanism can be exploited for arbitrary code execution with escalated privilege in the non-secure state by introducing a new exploitation technique, namely return-to-non-secure (ret2ns). To defend against ret2ns attacks, we design two address sanity checking mechanisms that verify the target address in the non-secure state is not unprivileged.

Sherloc

• Xi Tan
• Ziming Zhao

With the increasing deployment of the Internet of Things, their security is more important than ever. Many attacks, such as Return-Oriented Programming, aim to hijack the control flows of IoT programs.Control-flow Integrity is a promising approach to defend against such attacks. While existing solutions rely on instrumenting the code, which introduces extra execution complexity. In this work, we present Sherloc, which provides efficient control-flow violation detection to enhance the security of embedded systems without instrumentation. Sherloc combines static analysis on the target program's binary and run-time hardware support tracing mechanism to reconstruct run-time executions to detect control-flow hijackings. To protect the trace packets and detection engine, it utilizes TurstZone and places the target program in the non-secure state. We implemented and evaluated Sherloc on ARMv8-M architecture. Our evaluation shows Sherloc introduces a low overhead.

BYOTEE: Towards Building Your Own Trusted Execution Environments Using FPGA

• MD Armanuzzaman
• Ziming Zhao

Existing Trusted Execution Environments (TEE) or enclaves are built on proprietary hardware, such as Intel SGX and Arm TrustZone. Despite of the benefits, they have been criticized for lack of transparency, vulnerabilities, and various restrictions. For example, proprietary hardware only provide a static and fixed hardware Trusted Computing Base (TCB). Existing TEEs time-share a processor core with the Rich Execution Environment (REE), making execution less efficient and vulnerable to cache side-channel attacks. While previous projects attempted to address one or more issues by building software layers on SGX, TrustZone, or RISC-V, some issues are rooted in the design of hardware system; therefore they cannot be addressed by software alone. In this project, we present BYOTEE (Build Your Own Trusted Execution Environments), which is an easy-to-use hardware and software co-design infrastructure for building enclaves using Field Programmable Gate Arrays (FPGA). BYOTEE creates enclaves with customized hardware TCBs and establishes a dynamic root of trust that allows untampered execution of Security-Sensitive Applications (SSA) from preexisting software on the hardcore system. Additionally, BYOTEE provides mechanisms to attest the integrity of enclaves’ hardware and software stacks. We implement a BYOTEE system for the Xilinx System-on-Chip (SoC) FPGA. The evaluations on the low-end Zynq-7000 system for four SSAs and 12 benchmark applications demonstrate the usage, security, effectiveness, and performance of the BYOTEE framework.

Large-Scale Acoustic Automobile Fault Detection: Diagnosing Engines Through Sound

• Dennis Fedorishin

In this paper we present AMPNet, an acoustic abnormality detection model deployed at ACV Auctions to automatically identify engine faults of vehicles listed on the ACV Auctions platform. We investigate the problem of engine fault detection and discuss our approach of deep-learning based audio classification on a large-scale automobile dataset collected at ACV Auctions. Specifically, we discuss our data collection pipeline and its challenges, dataset preprocessing and training procedures, and deployment of our trained models into a production setting. We perform empirical evaluations of AMPNet and demonstrate that our framework is able to successfully capture various engine anomalies agnostic of vehicle type. Finally we demonstrate the effectiveness and impact of AMPNet in the real world, specifically showing a 20.85% reduction in vehicle arbitrations on ACV Auctions' live auction platform.

RidgeBase: Multifinger Cross Sensor Fingerprint matcher

• Bhavin Jawade
• Deen Dayal Mohan
• Srirangaraj Setlur
• Venu Govindaraju

RidgeBase consists of more than 15,000 contactless and contact-based fingerprint image pairs acquired from 88 individuals under different background and lighting conditions using two smartphone cameras and one flatbed contact sensor. RidgeBase is designed to promote research under different matching scenarios that include Single Finger Matching and Multi-Finger Matching for both contactless-to-contactless (CL2CL) and contact-to-contactless (C2CL) verification and identification. RidgeBase dataset can be used for training and evaluating contactless fingerprint matching algorithms (CL2CL and C2CL) for three types of tasks: Task 1: Single Finger Matching Task 2: Four Finger Matching Task 3: Set-Based Matching

Understanding and detecting IoT malware remote infection

• Qiqing Huang
• Hongxin Hu

The rocketed population of Internet of Things(IoT) devices has turned out to be ideal targets for attackers recently due to budget constraints and the enormous number of vulnerabilities with such IoT devices. Therefore, it is crucial to understand and detect the IoT malware remote infection before adversaries monetize the compromised IoT devices. This winning paper presents a study on a large-scale dataset covering 403,464 IoT malware samples with a deep insight into the IoT malware remote infection characteristics. Furthermore, to realize the early detection of remote infection, they developed an infection detector to monitor ongoing remote infection activities based on the infection capabilities. With deploying the infection detector on the software IoT honeypots for almost one year, they detected 147,860 remote infections with a 99.22% detection rate.

Understanding COVID19-Related Hateful Memes

• Name
• Email
• id

In this work, we propose three novel studies to understand V\&L models' generalizability for new types of hateful memes and the relationship between visual and textual modalities during prediction. To study generalizability, we use COVID-19 as a case study and conduct a large-scale measurement of state-of-the-art models to investigate their generalizability on COVID-19-related hateful memes. Furthermore, we conduct two novel experiments to understand the relationship between visual and textual modalities. Using gradient-based explanation techniques, we explore the importance attributed to each modality. Then, using ablations of the modalities, we scrutinize the impacts of textual and visual inputs on the models' prediction. Our studies show that these models are not generalizable to new types of hateful memes (average F1-score of only 0.2), give significantly more importance to the visual modality during prediction (about 2X higher average importance than textual modality), and that the textual modality actually provides significantly greater valid information to detect new types of hateful memes.

SCC-PG: A Sustainable and Connected Community-Scale Food System to Empower Consumers, Farmers, and Retailers

• Deen Dayal Mohan
• Bhavin Jawade
• Naji Mohamed Ali
• Rahul Nair

FoodNet is a community-powered social network built around a community's food system. FoodNet is a lightweight web application that helps community members locate inexpensive, fresh,local produce.

(A programming competition)

Low Power Object detection on FPGA system

• Pranay Meshram
• Naveena Elango
• Sheena Ratnam Priya
• Anthony DeMartino
• Yash Turkar
• Karthik Dantu
• Jinjun Xiong

A Low Power Object detection design on FPGA system with high accuracy and low power consumption. We worked on the problem statement of, how to design a lower power object detection AI algorithm that can be implemented and demonstrated on a piece of real computing hardware platform that can be attached to drones.

(CSE 611: Master's Capstone Project)

NetWorthy / Goally Finance

• Rahul Sharma
• Rahul Shahi
• Amol Gharpure
• Sumantra Sharma

Networthy is a personal finance startup focused on helping young alumni achieve financial wellness and freedom. This project involves building a web-based platform that allows customers (alumni) to list their financial goals, view related content (as per goal tags) provided by the coaches, work towards that goal, and progress through the goal stages.

Power2Peer

• Ashish Kumar

A Decentralized Marketplace for Solar Energy

The Food Court

• Gaoxiang Liu
• Yuanpu Zhang
• Ying Yin Li
• Hongyu Ke

The food court will be an app available on iOS and android. It will allow people to rediscover food, allowing them to seamlessly swipe through food, recipes and restaurants on the fly. The app will allow groups to combine their preferences to find something perfect to eat and cook.

Rentals&Friends

• Vanitha Ravichandran
• Indu Raju
• Manoj Chowdary Raavi
• Rakshokini Umasankar
• Hrisheekesh Talari

Rentals and friends is an application which aims to find rental apartments to people and as well as find tenants to sublease their apartments

Camo App (Maternal Health Project)

• Sai Meghana Muthyala
• Sravya Polina
• Sumana Radham
• Lokesh Chilamcharla
• Venkata Sai Dheeraj Narayanabhatla
• Sowbhagya Veera

Building a digital solution empowering birthing parents and caregivers to take charge of their prenatal and post-partum care.

Collaborative Clinical Tracker

• Sai Kumar Madhadi
• Vinay Kumar
• Sampath Bhargav Pinnam
• Rajesh Reddy Mekala
• Sonali Misra
• Rahul KaushikBhai Shah

Collaborative Clinical Tracker is a web application that can be used as a one-stop solution for students to log activities, track their progress, and generate reports of their training activities online, Instead of going through all the paperwork. It also helps department faculty and supervisors to easily access, monitor, and approve student training activities.

AFM-HD

• Bhavan Anand
• Deepak Gowda
• Shashank Yama
• Apoorva Bhapkar
• Sushmita Patil
• Sachin Ramesh Sarsambi

Building a web application for our client, which enables him to share his RV location and other events with his users and allow customers to book appointments for a haircut.

K-12 Q&A

• Sivakumar Pasupathi
• Vaishnavi Ruppa Gangadharan
• Sai Prashanth Selvaganapathy
• Brinda Ashar
• Parthiban Rajendran

A knowledge-sharing platform for K-12 Students/Faculty to learn and help others through various discussion threads

Capstone Project Management

• Ramesh Pavan Pothamsetty
• Pratik Chavan
• Shreya Anil Vinchankar
• Ashley Dsouza
• Nandakumar Shubhankar
• Siddharth Chillale

One stop portal for students to enroll into projects, for clients to manage their projects, judges to judge projects on any event and instructor to manage his courses, projects in courses and students.

Catalyst Ops

• Asheeque Chericham Veettil Maliyakkal
• Mathew Thengumpallil Jose
• Ann Gina Konnayil
• Ganesh Nayar Harindranathu
• Ritesh Manchikanti

This is an interactive Web Application with a smart AI chatbot that analyzes the profile requirements with user interaction. It generates curated search results of potential matches based on the conversational anchor points.

Find A Mechanic

• Yashwanth Gundela
• Roshan Sanjay Karamchandani
• Chitravardhini Bellamkonda
• Vaishnav Tammadwar

The project “Find a Mechanic” is a mobile application that will be used by the public to find a genuine and suitable car repair service with the requirements needed for the car owner nearby their location just by going through the app. As well as local repair services will also be able to promote and expand their business by the increase in the services and orders made through this app. This application will provide a lot of opportunities for local businesses and mechanics. And make it easier for car owners to repair their cars by finding a nearby mechanic with good skills and being able to check the status of their car.

Resume Parser and Intelligent Job Suggestions

• Manor Deshmukh
• Paritosh Katre
• Rahul Senapati
• Vidushi Sharma
• Harish Chandrasekaran

Our project will provide the functionality to create an applicant profile and upload the resume. The parser would then extract relevant information from the resume. Our matcher would subsequently try to match this extracted information with the job openings based on different matching metrics. Finally, the system will suggest suitable job openings to candidates based on insights from the matcher.

Dancestry

• 1. Sai Neeraj Sabbisetti
• 2. Girija Polamreddy
• 3. Gopi Chand Pendyala
• 4. Harinee Purushothaman
• 5. Srinivas Chaitanya Cheepuri
• 6. GOKUL SAI KADAPARTHI

Dancestry is an interactive, app-based lineal network illustrating connections between dance artists, their mentors and choreographers they have danced in the work of. Part social network and part archive, Dancestry collects, preserves, analyzes, and makes accessible dance lineage: who dancers have studied under and the choreographers whose work they have danced in. Dancestry is intended as a global resource for investigating artistic influences, career paths, choreographic connections and professional relationships.

NYCE

• Anirudh Sriram
• Josh Caskie
• Ethan Nhan
• Mukul Jeswani
• Michael Geraci

The NY Cannabis Exchange. An auction site built for the future of legal cannabis operations in New York.

Appoint-Mend

• Anuj Vadecha
• Harshitha Siddaramaiah
• Gautham Kidiyoor
• Vamsikrishna Murali

A Resource-based Appointment Scheduling system to be able to cater to all kinds of needs with fair logical seperation with administrative capabilities.

Fleet Tracker

• Srinand Andey
• Elijah Einstein
• Xiaobing Huang
• Saket Dixit

Fleet Tracker is a vehicle tracking system for car dealerships and their customers. Ordered cars can be tracked through location sensors fitted on them. We intend to access sensor data via the Helium blockchain API in future. Car dealerships can create an account on the app, add new vehicles, assign vehicles to customer email, assign devices to vehicles, view current location and location history of vehicles in their inventory. Customers may also view the live location of their car being delivered which facilitates smooth operation of the delivery.

Sonar Team

• Mitali Prabhakar Sable
• Gunjan Vivek Swamy
• Kartikeyan Lakshminarayanan Vignesh
• Lalitha Bhavani Palaparthi
• Chia Chen Chen

Sonar is a Chrome extension for students/professionals that returns read receipts for outgoing emails with the aim of making collaborative environments more efficient and updating an aging communication technology for a modern world.

(CSE 410: 3D Game Development in Unreal)

Project: Hunt

• Yiyuan Wang
• Jiacheng Wang

We made an action game with Unreal Engine 5.

The Draw

• Hannah Kloss
• Ria Gupta
• Alex Wright

"One Choice, Choose Wisely". A turn-based playing card game themed around the ancient Greek gods. Made in Unreal Engine 5.

The Legend of Pirate: Breath of the Sea

• Sungho Gold
• Haoxiang Lin
• Haohui Lin

This is a student group project video game created for the CSE 410 3D Game Development in Unreal course. The video game is an action hack-and-slash, and showcases the skills and progress we have made in the course.

Medusa Cafe

• Ruslana Savchuk
• Hala Alfadhli
• Abby Smardz
• and Susan Brethauer

Dating Sim/Visual Novel Game set in a cafe! Serve coffee to the gods and then maybe date them?!

TaxOfWar

• Joey Naval Poblete Jr
• James Christopher Aquilina III
• Rupin Raj Kumar Pradeep
• David Lam

Fantasy RPG

Turtle Life

• Joshua Arthur Jankowski
• Cindy J Cheng
• Thamara Ghislane Ferreira Gourgue
• Raven Tarbell

Turtle Life is a One-player adventure game. As a player you will follow the life of a pet turtle interacting with the environment to survive, explore, and ultimately escape the tank to the backyard pond. Throughout this journey, the turtle will be traveling through rooms across the house collecting strawberries and avoiding household obstacles such as a book maze, a roomba guarded kitchen, animals in the pond attacking an unfamiliar face, and of course the owner attempting to bring the turtle back home.

Kick Flicks

• Oakley Thomas
• Anthony Roberts
• Hema Bobba
• Sean Timmons
• Brett Sitzman

Based on the English tabletop soccer game “Subbuteo”, Kick Flicks is a physics based skill game that consists of two players going head to head in either a turn based or fast paced mode.

Abarkalypse

• Raymond Calascibetta
• Jake Kubiak
• Justin Kohn
• Danny Wu

Survive a zombie Apocalypse while searching for your lost companion who ran away when the virus outbreak began.

(CSE 546: Reinforcement Learning)

Air Traffic ContrRL

• Mason Lary
• Jue (Bob) Guo
• Yuting Hu

Millions of Americans travel by plane every day, and the safety of our airspace should be a top concern for all of us. Air Traffic Control is time-consuming and laborious due to the human element. Therefore, the automation of such a process is much needed. Utilizing advanced reinforcement learning methods (PPO), we have created a model which allows for automated traffic control of many agents so that they reach their destinations without conflict in a timely manner.

Comparing RL Methods in AWS DeepRacer

• Junqi Wu
• Amudheswaran Sankaranarayanan
• Abhilash Sampath

Amazon's AWS DeepRacer is a globally hosted competition where developers must program an RC racecar to complete laps around a circuit as fast as possible. The car is equipped with only a front-facing camera and no prior knowledge of the track. Reinforcement Learning (a branch of machine learning where agents are trained to take actions in an environment to maximize some reward) is used to tackle AWS DeepRacer. We have implemented several RL models using Soft Actor-Critic and Proximal Policy Optimization and compared them to a baseline from AWS. Our RC car is able to stay on the track at high speeds and complete a lap in under a minute.

Automated Stock Trading using Deep Reinforcement Learning

• Venkata Sai Kiran Yelchuri
• Debasmit Roy

Stock trading strategies play a critical role in investment. However, it is challenging to design a profitable strategy in a complex and dynamic stock market. In this project, we design strategies that employ deep reinforcement learning to maximize profit in a simulated day trading environment.

(CSE 546)

Solving complex driving scenarios using deep RL

• Abilash Chittrarasu
• Yeseswin Ashutosh Santhanam

The goal of the project is to aid both autonomous vehicles and drivers alike to make smart driving decisions which helps save time and impact the environment in a positive way by reducing carbon emissions(Traffic). We have used several real life driving scenarios and have trained a model to navigate through the given task in the most efficient way .

(CSE 546 : Reinforcement Learning )

ACCESSIBILITY IN REINFORCEMENT LEARNING

• Naman Khurpia
• Mohammed Junaid Shaik
• Komal Pulla

With the advent of self-learning and self- managing technologies, we choose to concentrate on the topic of using reinforcement learning for increasing accessibility for the specially abled. In this project we have considered a future where there are self-driving wheelchairs just like self-drive cars, and chose to explore the problem of comfort for the passenger in the wheelchair. We have used DQN to propose different routes for a person to reach a destination and help choose between reaching a place early or reaching a place comfortably.

(CSE 498TUT: Ug Research & Creative act)

VizierDB: A Reactive Data Science Notebook

• Maalolan Bharaniraj

FoodSystems project where we setup a data processing workflow starting with static sources and incorporate dynamic information. The purpose of this research is to facilitate Western New York population to be able to find nearest food resources with ease. Project is don using VizierDB a multi-lingual and multi-modal data centric notebook where spreadsheets, python and SQL are used for different purpose throughout the research.

(CSE 442: Software Engineering)

Look the Part/Fox in Socks

• Hope Kara
• Hannah Kloss
• Isabelle Ondracek
• Katherine Stock
• Hannah Wilcox

Fox in Socks developed an Android counter-part to the existing Look the Part iOS app. The app is for users interested in fashion seen in popular media, including TV shows.

Surgeon's Lair

• Nicholas Myers
• Austin Hockenberry
• Cindy Cheng
• Emily DeSantis
• Taisia Goncharouk

An educational game for medical students to diagnose damaged cranial nerves by simulating doctor and patient interactions in an emergency room.

Smart Mirror/jjank

• Kelly Chan
• Joyce He
• Junqi Wu
• Alexander Wang
• Nil Dhamecha

Developed a sunglasses module to use with smart mirror. Allows future consumers to shop and try on sunglasses products using augmented reality and machine learning. Created client facing UI where they are able to upload 2D images of sunglasses to the firebase database to allow consumers the different options to display on their face.

(CSE 616: Multiagent Systems)

Multi-Agent Systems Approach to Refugee Resettlement

• Fatima Umar

In recent years, governments and humanitarian agencies have started automating the refugee resettlement process. The problem is generally modeled as a bipartite matching between families and locations with two-sided preferences as well as demand and capacity constraints for a set of services. Researchers have proposed several algorithmic solutions, each claiming to improve efficiency or welfare in some aspect. The goal of this study is to evaluate and offer a comprehensive comparison of three such algorithms. Further, we show that each algorithm is susceptible to capacity manipulation, a form of strategic behavior where locations can improve their outcomes by misreporting their capacity constraints.