JIAN VORA

1029, Hostel 18, Indian Institute of Technology Bombay, Powai, Mumbai - 400076
® jianvora.github.io | [email protected] | ‡ jianvora

Research Interests
Statistical machine learning, Deep generative models, Probabilistic reasoning, Trustworthy machine learning
Education
Stanford University Stanford, CA, USA
Master of Science in Computer Science with specialization in Artificial Intelligence Beginning Sept 2021

Indian Institute of Technology Bombay Mumbai, MH, India

Bachelor of Technology in Electrical Engineering July 2017 – May 2021
• Dual Minor in Computer Science and Machine Intelligence & Data Science (C-MInDS)
• Major GPA : 9.77/10.0, C-MInDS Minor GPA: 10.0/10.0

Publications, Preprints, and Working Papers

4. Jian Vora, Isabel Valera, Guy Van den Broeck, Antonio Vergari, Play & Play Multimodal
Generative Model Allowing Tractable Inference, Working Paper

3. S. Jain, Jian Vora, I. Nair, D. Mehta, R. Shah, S. Gupta, S. Khyalia, S. Das, V. Riberio, S.
Kalyankrishnan, PAC Mode Estimation using PPR Martingale Confidence Sequences, under
review at 35th Conference on Advances in Neural Information Processing Systems (NeurIPS), 2021.

2. Jian Vora, Karthik S. Gurumoorthy, Ajit Rajwade, Recovery of Joint Probability Distribution
from One-Way marginals: Low rank Tensors and Random Projections, accepted at 2021 IEEE
Statistical Signal Processing (SSP) (SSP 2021).

1. Jian Vora and Ajit Rajwade, Compressive Signal Recovery Under Sensing Matrix Errors
Combined With Unknown Measurement Gains, accepted at ICASSP 2021 - 2021 IEEE
International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2021.

Research Experience and Internships

PAC Mode Estimation using PPR Martingale Confidence Sequences Jan 2021 – May 2021
Guide: Prof. Shivaram Kalyankrishnan, Department of Computer Science and Engineering IIT Bombay
• Used the prior-posterior ratio (PPR) as an α confidence sequence for mode estimation and performed simulations

with the application of predicting the results of an election with the help of polls
• Proved the sample complexity for the same and showed that it matches with the lower bound. Showed two

algorithmic versions - one vs one and one vs all and proved the optimality of the former for all problem instances
Low Rank Joint Probability Density Tensor Recovery from Marginals June 2020 – May 2021
Guide: Prof. Ajit Rajwade, Department of Computer Science and Engineering IIT Bombay
• Employed the canonical polyadic model for representation of a low rank joint probability density tensor and

formulated recovery of the same from marginals estimated from random projection of data
• Recovered the tensor using ideas from inverse radon transform, got better classification accuracies than standard

discriminative models and better performance when compared to other generative models
Multimodal Generative Modeling with Tractable Inference April 2020 – Ongoing
Guide: Prof. Guy Van den Broeck, Antonio Vergari, StarAI Lab University of California, Los Angeles
• Trained regularized autoencoders for each modality independently and employed a probabilistic circuit on the joint

latent space as a generative model allowing linear time exact inference in the size of the circuit
• Optimised the leaves to allow for heterogenous statistical data types and explored various schemes for end-to-end

training with the joint training of the autoencoders of all the modalities and the circuit on the latent space
• Explored various multimodal settings such as image-image(MNIST-SVHN), image-attributes(CelebA) and

image-text(CUB) getting better results over a variety of metrics over the baseline multimodal VAE models
 Efficient Learning of a Mixture of Log-Concave Densities Aug. 2020 – Jan. 2021
Guide: Prof. Vivek Borkar, Department of Electrical Engineering IIT Bombay
• Proved that random lower dimensional projections of mixtures of log concave densities is gaussian mixture model in

a total variation sense by using the intital result by Eldan and Klartag for a single isotropic log concave density
• Surveyed and implemented subspace clustering methods to find the directions where the densities concentrate

before fitting a GMM on the projected space followed by invoking the JL Lemma
Blind Calibration of Perturbations in Compressed Sensing Dec. 2019 – June 2020
Guide: Prof. Ajit Rajwade, Department of Computer Science and Engineering IIT Bombay
• Calibrated frequency offsets(gradient delays) and sensor gains(person in motion) in linear compressive measurement

framework which occur in MRI signal acquisition for faithful signal reconstruction using a single snapshot
• Proved uniqueness of the solution up to a scalar and obtained theoretical bounds on the error of signal

reconstruction by modelling the problem as a GMMV problem with additive signal dependent noise
Spatio-Temporal Action Detection and Classification in Videos May 2019 – July 2019
Guide: Dr. Martin Klinkigt, Hitachi Central Research Laboratory Tokyo, Japan
• The project was based on the TRECVid ActEV’19 challenge by NIST, proposed a stage-wise architecture of object

detection followed by tracking and activity classification on the VIRAT dataset

• Trained a deep-learning based Temporal Reasoning Network for activity classification achieving an accuracy of 95%;

tested the one-vs-all classification approach and designed an event finite-state machine for temporal alignment

Key Technical Projects

Continual Learning for Keyword Spotting | CS753: Automatic Speech Recognition Spring 21
• Implemented various regularizaton schemes like EWC, LWF, CRA for domain adaptation for keyword spotting

• Performed a joint keyword spotting and speaker verification task as a part of an Interspeech challenge

• Awarded one of the best projects award (top 4 out of 40+ projects)

Projection Efficient Sub-Gradient Descent | CS769: Optimization for Machine Learning Spring 21
• Surveyed literature to find about existing bounds for the number of function and projection orcales calls

• Implemented a recent NeurIPS paper on a low-rank SVM problem which acieves lower PO call complexity

StyleGAN for Music Generation | EE782: Advanced Machine Learning Fall 20

• Converted audio signal to an spectrogram using recent work on generative modelling of audio data such as

GanSynth and implemented a conditioning class feature in StyleGAN appended to the isotropic noise vector
• Modified the current StyleGAN architechture in order to allow for style transfer in audio signals

Shortest Path in a Maze | CS747: Foundations of Intelligent and Learning Agents Fall 20
• Modelled a 2D maze as an MDP with appropriate states, actions, rewards and transition probabilities

• Implemented various algorithms like Howard’s Policy Iteration, Value Iteration and Linear Programming to find the

best policy to be followed to minimize the number of steps between 2 given points while following the constraints

Iris Recognition and Classification | CS663: Digital Image Processing Fall 19

• Segmented pupil and iris from a dataset of eyes using binary thresholding and circular hough transform

• Implemented PCA for dimensionality reduction followed by LDA for classification achieving an accuracy of 86%

Multicycle & Pipelined RISC Processors | EE309: Microcontrollers, EE224: Digital Systems Fall 19
• Designed and implemented 6-stage pipelined and multicycle RISC processors in VHDL, consisting a total of 16

arithmetic,logical and branching instructions, and tested it on Altera DE0-Nano FPGA board
• Developed a robust model to handle hazards by integrating data forwarding and stalling units to reduce latencies

Technical Skills
Programming Languages: C++, Python, VHDL, MATLAB, Bash, HTML, Embedded C
Machine Learning: PyTorch, TensorFlow, Caffe, OpenCV, Numpy, Pandas, Matplotlib
Software: Docker, GNUPlot, LATEX, Git, Scilab, SolidWorks, NGSpice, Arduino, Quartus, Modelsim
Major Honors and Awards
• Awarded an option of branch change due to exceptional academic performance in the first year
• Selected for undergraduate research internship programs at Caltech (SURF) and NTU India Connect
• Achieved All India Rank 3 in ICSE 2015 out of 0.16 Million applicants
• Secured All India Rank 645 in JEE Advanced 2017 among 0.17 Million candidates
• Bagged the silver medal in Homi Bhabha Balvaidnyanik (top 0.4%) search for scientific and research aptitude
• Awarded the Inspire Scholarship by Govt. of Maharashtra for being in the top 1% in Class XII Examinations

Relevant Coursework
Computer Science: Foundations of Learning and Intelligent Agents, Introduction to Machine Learning, Advanced
Machine Learning, Theoretical Machine Learning, Data Structures and Algorithms, Digital & Advanced Image
Processing, Cryptography and Network Security, Advances in Learning and Intelligent Agents, Automatic Speech
Recognition, Operating Systems, Science of Information Learning and Statistics
Mathematics and Statistics: Probability and Random Processes, Data analysis, Linear Algebra, Calculus, Ordinary
and Partial Differential equations, Complex analysis, Optimization for Machine Learning
Electrical Engineering: Signals and Systems, Control Theory, Analog and Digital Circuits, Communication Systems,
Digital Signal Processing, Microcontrollers, Network Theory, Power Electronics, Electronic Devices

Teaching, Mentoring and Leadership Roles

Teaching Assistant | MA 111, Vector Calculus Jan 2021 – Mar 2021
• Responsible for taking weekly tutorial (problem solving) sessions for a batch of 50 students

• Involved in setting and correcting examinations and deciding logistics for an online mode of the course

Manager | Electronics and Robotics Club, IIT Bombay Apr 2019 – Apr 2020
• Led a team of 17 members to boost the Electronics and Robotics culture across the institute

• Initiated and maintained a reading group VisionX for discussing and implementing recent research in the field on

computer vision with members including professors, graduate and senior undergraduate students
• Conducted workshops and hackathons by various companies like Uber, Intel and Texas Instruments

• Mentored 3 month long technical summer projects taken up by freshmen and helped in their ideation

Institute & Department Academic Mentor | Student Mentorship Program Apr 2019 – May 2021
• Responsible for mentoring a group of 12 freshmen students to ensure their academic and holistic development

• Responsible for co-ordinating with the department as a senior DAMP mentor, organised a lecture on mental health

awareness and stress management, mentoring students to clear off their academic backlogs

Extra Curricular Activities

• Worked with NGO Asha , designed and taught a basic computer and English speaking course for on campus workers
to enable them to be eligible for entry level jobs which require these basic skills
• Mentored 8 students under Summer of Science by providing them necessary guidance in machine learning; spoke at a
15 day long Machine Learning bootcamp with 250+ participants and developed assignments and projects
• Stood 1st among 25 groups in MHRD-TEQIP-3 Activity which involved presenting the topic “Mathematics in
Engineering” to professors and students of other universities
• Represented Maharashtra in a quiz competition with the theme of Indian history at the national level