Portfolio
More on my personal background: I am a U.S. citizen who grew up and was educated in Singapore. Besides research, I worked as a software engineer at Salesforce and as a product strategist at several startups, ranging from deep tech (quantum computing) to B2C companies based in Southeast Asia and the United States. I also studied abroad at the University of Oxford (Magdalen College) in Fall 2022, where I studied graph representation learning and philosophy of mind, and tried my hand at rowing! I enjoy playing tennis, snowboarding, ashtanga yoga, hiking, reading (+ occasionally writing) science fiction, and brush calligraphy.
Inspired by my interdisciplinary coursework, I am drawn to research leveraging cognitive science for robot learning and visual understanding. I aim to better understand human cognitive processes, such as multimodal perception, curiosity, and interactive learning, to develop human-inspired learning algorithms for robotics.
Below is an assortment of works that summarize my academic interests.
Thesis Papers
Leveraging Affordance Representations for Robot Learning [thesis] [publication]
Undergraduate Honors Thesis
Humans are capable of adapting to novel environments quickly using prior knowledge from past experiences. We can identify new instantiations of previously encountered object classes and easily apply previously learned skills to these new objects, both of which current embodied AI agents struggle with. Online reinforcement learning, where a robotic agent learns a mapping from states to actions to maximize a reward signal, provides a potential solution by enabling robots to learn from trial-and-error. However, current methods are sample-inefficient, lack shaping rewards, and require frequent resets. We propose a method to address the lack of shaping rewards using affordances, the action potential of objects, to create a dense shaping reward for online reinforcement learning. We leverage state-of-the-art vision-language models (VLMs) to predict keypoint-based affordance representations, which we use as intermediate dense rewards for online reinforcement learning, in addition to sparse task completion rewards. We demonstrate that dense shaping rewards speed up online reinforcement learning for robotic manipulation, and enables robots to succeed on a variety of object manipulation tasks, informed by human interaction priors encoded in VLMs.
Topics: Affordances, Online Reinforcement Learning, Vision-Language Models, Robotics, Learning from Human Videos
Teaching
Summer 2024: CS 229 Machine Learning
Taught by Prof. Jehangir Amjad
Topics: Machine Learning, Supervised Learning, Unsupervised Learning
Winter 2024, Spring 2024: CS 224N Natural Language Processsing with Deep Learning
Taught by Prof. Tatsunori Hashimoto / Prof. Diyi Yang (Winter 2024) and Prof. Christopher Manning (Spring 2024)
Topics: Natural Language Processing, Machine Learning, Deep Learning
Fall 2023, Fall 2024: CS 157 Computational Logic
Taught by Prof. Michael Genesereth
Topics: Propositional Logic, Relational Logic, Functional Logic
Computational Projects
Today Years Old: Adapting Language Models to Word Shifts [paper] [poster] [code]
Final report, poster, and code for Stanford’s CS 224N: Natural Language Processing with Deep Learning (Winter 2023)
Finetuned GPT-2 and RoBERTa to predict word embeddings for novel lexical items from Urban Dictionary given their definitions.
Topics: Natural Language Processing, Machine Learning, Supervised Learning, Domain Adaptation
A Shot in the Dark: Modeling Improved Zero-Shot and Few-Shot Transfer Learning with Self-Supervised Models for Sentiment Classification [paper] [poster]
Final report and poster for Stanford’s CS 229: Machine Learning (Spring 2022)
Modeled transfer learning with self-supervised embeddings to optimize model performance on sentiment classification tasks.
Topics: Natural Language Processing, Machine Learning, Self-Supervised Learning, Transfer Learning
Model Predictive Curiosity [paper] [poster]
Final report and poster for Stanford’s PSYCH 240A: Curiosity in Artificial Intelligence (Spring 2022)
Proposed Model Predictive Curiosity (MPCu) to optimize for high-curiosity action values and enrich multi-object interactions in a Box2D environment.
Topics: Curiosity-Based Models, Model-Based Reinforcement Learning, Representation Learning, Self-Supervised Learning
Machine Learning-based platform using iBeacon Sensors for Product Location and Indoor Navigation to Improve Consumer Retail Experience
High school research engineering project (2018-2019)
Trained an automatic speech recognition engine contextualized to Singaporean accents and terminology. Created a mobile app to help consumers navigate local supermarkets with verbal queries.
Topics: Natural Language Processing, Speech Recognition, Speech-To-Text, Recommendation Systems, Shortest Path Generation, Indoor Geolocation, Bluetooth Sensor Systems
Philosophy Papers
The Missing Piece: Dispelling the Mystery of Introspective Illusion [paper]
Final paper for Stanford’s PHIL 186: Philosophy of Mind (Spring 2023)
I argue that explaining the potency of phenomenal illusions is the crucial missing piece for a sound illusionist theory. I present two main desiderata for a positive theory of illusionism by drawing connections to related theories of consciousness, namely global workspace theory (Dennett, 2001) and Buddhist philosophy.
Topics: Consciousness, Illusionism, Higher-Order Thought Theory, Wittgenstein
Consciousness, Phenomenality, and the Representational Layer [paper]
Final paper for Stanford’s SYMSYS 202: Theories of Consciousness (Winter 2023)
I propose that metacognition on top of the representational layer, beyond mere possession of representational states, is critical for consciousness, and explore how phenomenality is introduced in this process. I discuss the implications of this proposal for the functional and evolutionary roles of consciousness.
Topics: Consciousness, Representationalism, Phenomenality, Higher-Order Thought Theory, Global Workspace Theory
Philosophy of Mind: Wittgenstein, The Unconscious Mind, and Self-Knowledge [paper]
Collection of essays for OSPOXFRD 199: Philosophy of Mind (Fall 2022)
A compilation of essays written for my Directed Reading in Philosophy of Mind, during my quarter abroad at Oxford. Each essay was written to prepare for biweekly tutorial discussions over the quarter. Topics included other minds, the privacy of experience, the unconscious mind, and self-knowledge.
Topics: Philosophy of Mind, Philosophy of Psychology, Wittgenstein, Private Language Argument, Consciousness, Self-Knowledge
Predictive Processing: Efficiently processing high-dimensional, multimodal inputs [paper]
Final paper for Stanford’s SYMSYS 205: The Philosophy and Science of Perception (Spring 2022)
I explore the plausibility of the predictive processing framework over the standard bottom-up model of perception. I specifically explore efficient processing of high-dimensional multimodal inputs, where the qualitative space of each modality has unique dimensionality and structure.
Topics: Multimodal Perception, Perceptual Cognition, Cognitive Processing
Large Language Models: Intelligence, Understanding, and Intentionality [paper]
Final paper for Stanford’s SYMSYS 207: Conceptual Issues in Cognitive Neuroscience (Fall 2021)
I argue that modern large language models (LLMs) cannot achieve strong intelligence.
Author’s Note (March 2023): This paper was written before the release of ChatGPT and GPT-4 (or GPT-x, depending on how far in the future you’re reading this). In hindsight, I acknowledge this paper does not give sufficient credit to the impressive emergent behaviors observed in LLMs. However, my stance towards purely language-based models are still generally aligned with this paper. Another work that articulates views I am sympathetic to is Shanahan (2022). That said, there are many cool developments expanding on LLMs (like vision-language models, or grounded language models more generally) that I’m excited about!
Topics: Natural Language Processing, Artificial Intelligence, Natural Language Understanding, Intentionality
Mathematics Papers
Asymmetric Processes [paper]
Research paper for Stanford’s MATH 101: Math Discovery Lab (Winter 2024)
Analyzes two asymmetric Markov models: first where particle number stays constant, and second where particles enter and exit at certain rates. We study probability distributions of particle configurations at equilibrium, and properties such as average particle speed, irreducibility, aperiodicity, and double stochasticity.
Topics: Markov Processes, Markov Chains, Probability Theory
Infinite Coin Tosses [paper]
Research paper for Stanford’s MATH 101: Math Discovery Lab (Winter 2024)
Explores cumulative distribution functions for infinite coin tosses, parameterized by the probability p of flipping heads. We graph the outcomes of simulated coin flips and study properties of the cumulative distribution function, analyzing its pathological behavior in terms of continuity, differentiability, and arc length.
Topics: Probability Theory, Continuous Random Variables
Hilbert’s 10th Problem [paper]
Research paper for Stanford’s PHIL 152: Computability and Logic (Spring 2023)
A proof of Hilbert’s 10th Problem: determining the solvability of Diophantine equations over integers.
Topics: Computability Theory
Technical Reports
The Future of Human-Machine Interaction: Keeping Humans in the Loop [paper]
Final paper for Stanford’s OSPOXFRD 29: Artificial Intelligence & Society (Fall 2022)
The doomsday ending that humans will be demolished in the fierce intelligence competition with AI systems, while remarkably enduring, is a narrow view that distracts us from active measures that can be taken in the present day. I assert that a key tenet of AI development going forward should be keeping humans in the loop,, and evaluate three technical research areas facilitating human-in-the-loop AI development and deployment. Distinguishing between non-immediate decision making (e.g., data analytics and robotics) and time-sensitive, safety-critical decision making (e.g., autonomous vehicles and aircraft) is key to understanding how to best facilitate human-AI collaboration in each case.
Topics: Human-AI Interaction, AI Safety, Human-In-The-Loop Development, Decision-Making
A Proposal for Building Safety Benchmarking Services in CAIS systems [paper]
Final report for Stanford Existential Risk Initiative’s AI research program (Spring 2021)
I propose a protocol encompassing safety benchmarking services for CAIS systems, ranging from pre-deployment safety benchmarks applied during model training to post-deployment safety benchmarks.
Topics: AI Safety, Benchmarking Tools, AI Existential Risk
Implications of the Comprehensive AI Services Framework on AI Safety Research [paper]
Final report for Stanford Existential Risk Initiative’s AI research program (Winter 2021)
I argue that developing powerful AI systems in line with the Comprehensive AI Systems (CAIS) framework outlined in Reframing Superintelligence (2019) should be encouraged, due to the potential for enhanced safety measures to mitigate AI existential risk.
Topics: AI Safety, Hierarchical Reinforcement Learning, AI Existential Risk
Autonomous Vehicles: From Vision to Reality [paper]
Final paper for Stanford’s CS 56N: Great Discoveries and Inventions in Computing, taught by Prof. John Hennessey (Fall 2020)
We provide an analysis of current developments in autonomous driving, and discuss the technological hurdles that lie ahead in enhancing the security of autonomous systems and implementation of the system at scale.
Topics: Autononous Driving, Computer Vision, LiDAR/RADAR Sensor Systems, AI Safety