Favyen Bastani

Contact: favyenb@allenai.org
Github: uakfdotb
CV/Resume: pdf

About: (Update: I am now an applied research scientist at the Allen Institute for AI.) I am a PhD student in computer science at the Massachusetts Institute of Technology working with Sam Madden in the data systems research group.

Research: Most recently, my research interests have revolved around the broad goal of empowering more users to be able to leverage machine learning techniques to analyze video. Commercialized or otherwise deployed use cases of ML in video analytics today are typically focused on a narrow range of applications, often of questionable utility, such as privacy-intrusive surveillance and facial recognition. My (perhaps misguided) hope is that, by reducing the effort and expertise needed to implement ML pipelines, we can enable more applications that are beneficial (or at least not detrimental) to society.

Here are some examples of applications of computer vision which people are in fact actively working on that I find compelling, which I think showcase the diversity of what's possible:

• Cornell's Bird Cams Lab is exploring opportunities in ecology, e.g., finding altercation events in bird feeder video where birds fight
• MIT's AgeLab is studying how drivers interact with vehicle automation (such as lane keeping and automatic emergency braking) to understand the impact of this technology and how it can best be used to enhance safety
• Various city and state governments, such as the Arizona DOT, are deploying traffic cameras to infer traffic volume and traffic safety metrics to inform both real-time and long-term planning decisions; these stakeholders generally have a strong interest in privacy (e.g., scrubbing video to erase pedestrians, or computing metrics without storing video)
• Autonomous robotics; including applications for the search part of search-and-rescue, and space exploration (e.g., Mars Perseverance)
• The Cable TV News Analyzer from Stanford's Computer Graphics Lab supports the "large-scale, data-driven analysis of the contents of cable TV news"

Under this goal, my general research interests include:

• Building systems that make it easier for users with little expertise to develop ML pipelines for video analytics (see Vaas, SkyhookML)
• Designing optimization techniques that accelerate the execution of ML pipelines on large datasets without impairing accuracy (see MIRIS, OTIF)
• Exploring use cases for self-supervisory signals in training ML models, which can reduce the amount of data needed to attain high-accuracy (see self-supervised MOT)

Publications

OTIF: Efficient Tracker Pre-processing over Large Video Datasets. Favyen Bastani, Sam Madden. SIGMOD 2022. [pdf] [web]

Self-Supervised Multi-Object Tracking with Cross-Input Consistency. Favyen Bastani, Songtao He, Sam Madden. NeurIPS 2021. [pdf] [web]

SkyQuery: An Aerial Drone Video Sensing Platform. Favyen Bastani, Songtao He, Ziwen Jiang, Osbert Bastani, Sam Madden. Onward! 2021. [pdf] [web]

Updating Street Maps using Changes Detected in Satellite Imagery. Favyen Bastani, Songtao He, Satvat Jagwani, Mohammad Alizadeh, Hari Balakrishnan, Sanjay Chawla, Sam Madden, Mohammad Amin Sadeghi. SIGSPATIAL 2021. [pdf] [web]

Beyond Road Extraction: A Dataset for Map Update using Aerial Images. Favyen Bastani, Sam Madden. ICCV 2021. [pdf] [web]

Vaas: Video Analytics at Scale. Favyen Bastani, Oscar Moll, Sam Madden. VLDB 2020, Demonstration. [web]

MIRIS: Fast Object Track Queries in Video. Favyen Bastani, Songtao He, Arjun Balasingam, Karthik Gopalakrishnan, Mohammad Alizadeh, Hari Balakrishnan, Michael Cafarella, Tim Kraska, Sam Madden. SIGMOD 2020. [pdf] [web]

Sat2Graph: Road Graph Extraction through Graph-Tensor Encoding. Songtao He, Favyen Bastani, Satvat Jagwani, Mohammad Alizadeh, Hari Balakrishnan, Sanjay Chawla, Mohamed Elshrif, Sam Madden, Mohammad Amin Sadeghi. ECCV 2020. [arxiv]

BeeCluster: Drone Orchestration via Predictive Optimization. Songtao He, Favyen Bastani, Arjun Balasingam, Karthik Gopalakrishnan, Ziwen Jiang, Mohammad Alizadeh, Hari Balakrishnan, Michael Cafarella, Tim Kraska, Sam Madden. MobiSys 2020. [pdf] [web]

RoadTagger: Robust Road Attribute Inference with Graph Neural Networks. Songtao He, Favyen Bastani, Satvat Jagwani, Edward Park, Sofiane Abbar, Mohammad Alizadeh, Hari Balakrishnan, Sanjay Chawla, Sam Madden, Mohammad Amin Sadeghi. AAAI 2020. [arxiv]

Machine-Assisted Map Editing. Favyen Bastani, Songtao He, Sofiane Abbar, Mohammad Alizadeh, Hari Balakrishnan, Sanjay Chawla, Sam Madden. SIGSPATIAL 2018. [pdf] [web] [code]

RoadTracer: Automatic Extraction of Road Networks from Aerial Images. Favyen Bastani, Songtao He, Sofiane Abbar, Mohammad Alizadeh, Hari Balakrishnan, Sanjay Chawla, Sam Madden, David DeWitt. CVPR 2018. [pdf] [web] [code]

RoadRunner: Improving the Precision of Road Network Inference from GPS Trajectories. Songtao He, Favyen Bastani, Sofiane Abbar, Mohammad Alizadeh, Hari Balakrishnan, Sanjay Chawla, Sam Madden. SIGSPATIAL 2018. [pdf] [web]

Large Scale Real-time Ridesharing with Service Guarantee on Road Networks. Yan Huang, Favyen Bastani, Ruoming Jin, Xiaoyang Sean Wang. VLDB 2014. [pdf] [code]

A Greener Transportation Mode: Flexible Routes Discovery from GPS Trajectory Data. Favyen Bastani, Yan Huang, Xing Xie, Jason Powell. SIGSPATIAL 2011. [pdf]

Code

SkyhookML is a platform for computer vision, providing an easy-to-use interface that makes machine learning methods for image and video data more accessible.

Vaas is an interactive web application for efficiently analyzing large-scale video datasets. It incorporates tools to annotate video, train machine learning models, and build data flow graphs that implement analytics tasks by composing diverse ML and pre/post-processing operations. See the project website for more details.

RoadTracer aims to improve the accuracy of inferring road networks from satellite imagery through an iterative graph construction procedure. The code is available in Github, with an improved version available here.

gobearmon is a simple but reliable distributed uptime monitoring system that can send alerts over e-mail and SMS when services go offline.