Hi! My name is Jeffrey and welcome to my blog and website!
I am an incoming software engineer at Crusoe Energy. I completed my M.S and B.S in Electrical & Computer Engineering at Carnegie Mellon in 2022 and 2021 respectively. During this time I focused primarily on deep learning and computer architecture/systems.
Currently my interests lie in deep learning, and the intersection of machine learning and computer systems. My research interests center around applying biologically inspired ideas to deep learning.
For fun, I love teaching, playing and watching sports (hockey, tennis, sport climbin, golf), and hanging out with friends. I'm also trying to blog about personal milestones and instructive academic content.
HighMMT: Quantifying Modality & Interaction Heterogeneity for High-Modality Representation LearningPaul Pu Liang , Yiwei Lyu , Xiang Fan , Jeffrey Tsaw , Yudong Liu , Shentong Mo , Dani Yogatama , Louis-Philippe Morency , Ruslan Salakhutdinov Submitted to TMLR 2022 An information theory-centric metric quantifying modality heterogeneity is introduced to measure affinity between modalities across tasks. This metric is used to inform parameter groupings in a Transformer based multimodal multitask model that achieves SOTA across several tasks with a significant reduction in parameters |
Hybrid Neural Model for Solving Math Word ProblemsProject for 10-707 Advanced Deep Learning | Spring 2022With Raghav Bansal We combined SOTA bottom up MWP solver with a SOTA top-down MWP solver into a novel hybrid model. Developed a novel query network to substitute probable subtrees from bottom up network with top down network's goal vectors. Achieved 75% theoretical accuracy on Math23k beating previous SOTA of 74% |
|
Autovöt: An Autonomous RC Vehicle ConvoyUndergraduate Capstone Project | Spring 2021With Joel Anyanti, Fausto Leyva Final Paper | Video | Poster We developed an RC vehicle convoy from scratch capable of autonomously navigating an obstacle course. Lead vehicle uses a perception stack and SLAM to send directions via bluetooth to following vehicle to navigate obstacles. Won 1st Runner Up out of 40+ projects in Spring 2021 |
|
RISCV ProcessorSolo Project for 18-447 Intro to Computer Architecture | Spring 2021I designed and wrote a synthesizable superscalar out-of-order pipelined RISCV processor in SystemVerilog that implements a subset of the RV32I ISA. Implemented optimizations such as a BTB, instruction stall buffer, and pipeline flushes. Achieved 280 MIPS on perscribed benchmarks and 1st Quartile performance out of 20+ teams. |
|
Recurrent GANs for Music GenerationFinal project for 10-701 Introduction to Machine Learning (PhD) | Fall 2020With Azaan Rehman We extended a baseline recurrent BiLSTM GAN for music generation and developed a novel architecture containing an input mapping network, convolution and attention layers, and a peripheral classifier using Tensorflow v1 in an AWS EC2 instance to generate statistically more realistic classical music compared to the baseline. |
|
Power Grid RL OperatorSummer 2020 I implemented and trained a DQN and a Double Duelling DQN with PER agents from scratch on Tensorflow 2.0 to achieve 3500% increase in operation before failure and an 11000% increase in aggregate reward per episode on grid2op environment compared with the baseline agents. |
|
Resume-Job Compatability AlgorithmWith Samantha Ramsey and Shaan Dave Winner of Credit Suisse Coding Challenge 2019 | Summer 2019 We designed an algorithm for picking the most qualified resume based on a job description through an NER model and a matching algorithm to compute an aggregate score. We also identified the most suitable job given a resume using word2vec with cosine similarity. |
|
The Empathetic JukeboxFinal Project for 15-112 Fundamentals of Programming and CS | Spring 2018I trained an emotion classifier with OpenCV and generated Spotify playlists based on the user's emotion using the Spotify API, and plays the songs automatically in real-time by webscraping YouTube. |
Apple Inc.GPU Hardware Intern (virtual) | Fall 2021 I worked on GPU memory verification team. Integrated idle checks and developed coverage for 5 modules within GPU routing block. Developed novel strategy to hit previously un-hit coverage points using Xceligen ML tool to improve coverage of a coverpoint by over 10% |
|
Credit SuisseTechnology Analyst Intern (virtual) | Summer 2020 I interned with Global Markets Data Analytics and developed a 2-stage pipelined model to extract bond information from Bloomberg chat data in Python in an Agile development environment. |
18-100 Introduction to Electrical and Computer EngineeringTA | Fall 2021 - Fall 2022 I taught recitation, held office hours and small groups, and wrote exam and homework problems on a variety of concepts including digital logic, transistors, op amps, signal processing and filters, and machine learning. |
|
10-701 Introduction to Machine Learning (PhD)TA | Spring 2021 I taught recitation, held office hours, and wrote exam and homework problems on various ML concepts, including regression and optimization, deep learning, reinforcement learning, graphical models and learning theory. |
|
18-349 Introduction to Embedded SystemsTA | Fall 2020 I held weekly small group sessions and office hours teaching fundamental concepts in embedded systems including ARM Thumbv2, serial protocols, timers and interrupts, threading and scheduling, networking, and security |
|
33-141 Physics I for Engineering StudentsSI Leader | Fall 2019 I organized and led biweekly sessions of 30+ students, as well as several exam review of 100+ students to review lecture material on kinematics, mechanics, energy and momentum, and thermodynamics. I also wrote and prepared problem sets and solutions for each session and exam review. |
|
15-213 Introduction to Computer SystemsTA | Summer 2019 I led lecture activities and held office hours on content covering system-level I/O, network protocols, bit-level representations, x86-64 programming, exceptional control flow (processes and signals), virtual memory, and caching. I also developed and benchmarked test traces to stress students' implementations of malloc memory allocator. |
|
33-142 Physics II for Engineering StudentsEXCEL Leader | Spring 2019 I organized and led 3 weekly sessions of 9-10 students, covering material on electric fields, Gauss' law, magnetism, electromagnetic induction, and Maxwell's Equations. I also wrote and prepared problem sets for each session, and co-developed and co-led multiple exam reviews with 100+ students. |
|
21-259 Calculus in Three DimensionsEXCEL Leader | Fall 2019 I organized and led 3 weekly sessions of 9-10 students, covering material on multivariable limits, differentiation, integration, surface and line integrals, and Stoke's and Divergence Theorems. I also wrote and prepared problem sets for each session, and led 3 exam reviews with up to 30 students. |
|
33-141 Physics I for Engineering StudentsTAA | Fall 2019 I co-led biweekly recitations for 25+ undergraduate students, providing support on concepts including kinematics, mechanics, energy and momentum, and thermodynamics |
Note: Spring 2020 Semester was completed abroad at UCL, and Fall 2021 was a leave of absent for an internship at Apple.