Alexander C. Li

I'm a second year PhD student in the Machine Learning Department at Carnegie Mellon University. I'm advised by Deepak Pathak, and my work is supported by the National Science Foundation Graduate Research Fellowship.

Previously, I was an electrical engineering and computer science major at UC Berkeley, where I received my BS and MS. I was a researcher in the Berkeley Artificial Intelligence Research Lab and was advised by Pieter Abbeel, Lerrel Pinto, and Carlos Florensa.

Email  /  CV  /  Google Scholar  /  LinkedIn  /  Twitter

I'm interested in self-supervised learning, generalization, optimization, and the role of data in deep learning.

We empirically analyze non-contrastive self-supervised methods and find that SimSiam is extraordinarily sensitive to model size. In particular, SimSiam representations undergo partial dimensional collapse if the model is too small relative to the dataset size. We propose a metric to measure the degree of this collapse and show that it can be used to forecast the downstream task performance without any fine-tuning or labels. We further analyze architectural design choices and their effect on the downstream performance. Finally, we demonstrate that shifting to a continual learning setting acts as a regularizer and prevents collapse, and a hybrid of continual and multi-epoch training can improve linear probe accuracy by up to 18 percentage points with ResNet-18 on ImageNet.

We propose a simple architecture for deep reinforcement learning that can control how quickly the network fits different frequencies in the training data. We explain this behavior using infinite-width analysis with the Neural Tangent Kernel, and use this to prioritize learning low-frequency functions and speed up learning by reducing networks' susceptibility to noise in the optimization process, such as during Bellman updates. Experiments on standard state-based and image-based RL benchmarks show clear sample-efficiency gains, as well as increased robustness to added bootstrap noise.

We present Generalized Hindsight: an approximate inverse reinforcement learning technique for relabeling behaviors with the right tasks. Given a behavior generated under one task, Generalized Hindsight finds a different task that the behavior is better suited for. Relabeling a trajectory with this different task and training with an off-policy RL algorithm improves sample-efficiency and asymptotic performance on a suite of multi-task navigation and manipulation tasks.

We develop a new hierarchical RL algorithm that can efficiently adapt pre-trained skills on related tasks, and directly learn effective emergent skills by simultaneously training the entire hierarchy.

We attempt to use powerful autoregressive models for image-to-image translation and outlier detection, with poor results. We analyze the failure modes and find that (a) differentiating through these models creates an ill-conditioned optimization problem and (b) their density estimates are incredibly sensitive to distributional shifts.

We develop a novel algorithm for estimating sunspot rotation speed that is more robust to feature evolution than traditional methods based on time-slices. We use this technique to gain more accurate estimates of magnetic energy buildup in rotating sunspots.

• National Science Foundation Graduate Research Fellowship (2020)
• 1st place at Citadel San Francisco Invitational Data Open (2019)
• Mark D. Weiser Excellence in Computing Scholarship (2019)
• Accel Scholar (2018)
• Quantedge Award for Academic Excellence: awarded to ~120 seniors at UC Berkeley for a 4.0 GPA (2018)
• Cal Alumni Association Leadership Award (2017)
• Dean's Honor List: Fall 2016, Spring 2017, Fall 2017, Spring 2018, Spring 2019
• Robert J. Kraft Award for Freshmen (2016)
• Regents’ and Chancellor’s Scholar: awarded to top <2% of entering students at Berkeley (2016)

Website template from Jon Barron.