- talks - lists.cs.princeton.edu

Samyak Gupta will present his General Exam on April 15th, 2022 at 1pm via Zoom.
by Nicki Mahler 12 Apr '22

12 Apr '22

Samyak Gupta will present his General Exam on April 15th, 2022 at 1pm via Zoom. Zoom link: [ https://princeton.zoom.us/my/samyakg | https://princeton.zoom.us/my/samyakg ] The members of his committee are as follows: Kai Li (adviser), Danqi Chen, and Jon Cohen (Neuroscience) Link to Abstract and Reading list: [ https://docs.google.com/document/d/1t6BU6TflwycaOe7XECeUk7Tln19Aui9JqVOLhxk… | https://docs.google.com/document/d/1t6BU6TflwycaOe7XECeUk7Tln19Aui9JqVOLhxk… ]

1 1

Vishvak Murahari will present his General Exam "DataMUX: Data Multiplexing for Neural Networks" on Monday, April 18, 2022 at 3:30 pm in CS 302 and via Zoom.
by Louis W. Riehl 11 Apr '22

11 Apr '22

Vishvak Murahari will present his General Exam "DataMUX: Data Multiplexing for Neural Networks" on Monday, April 18, 2022 at 3:30 pm in CS 302 and via Zoom. Zoom Link: [ https://princeton.zoom.us/j/3829472199 | https://princeton.zoom.us/j/3829472199 ] Committee Members: Karthik Narasimhan (adviser), Danqi Chen, Sanjeev Arora Abstract: In this talk, we introduce data multiplexing (DataMUX), a technique that enables deep neural networks to process multiple inputs simultaneously using a single compact representation. DataMUX demonstrates that neural networks are capable of generating accurate predictions over mixtures of inputs, resulting in increased throughput with minimal extra memory requirements. Our approach uses two key components -- 1) a multiplexing layer that performs a fixed linear transformation to each input before combining them to create a mixed representation of the same size as a single input, which is then processed by the base network, and 2) a demultiplexing layer that converts the base network's output back into independent representations before producing predictions for each input. We show the viability of DataMUX for different architectures (Transformers, and to a lesser extent MLPs and CNNs) across six different tasks spanning sentence classification, named entity recognition and image classification. For instance, DataMUX for Transformers can multiplex up to 20x/40x inputs, achieving 11x/18x increase in throughput with minimal absolute performance drops of <2% and 4% respectively on MNLI, a natural language inference task. We also provide a theoretical construction for multiplexing in self-attention networks and analyze the effect of various design elements in DataMUX. Reading List: [ https://docs.google.com/document/d/1Pi-Pd7NwcScb1UfbHaGaE_g_-6-cwqXM_18yQ_u… | https://docs.google.com/document/d/1Pi-Pd7NwcScb1UfbHaGaE_g_-6-cwqXM_18yQ_u… ] Everyone is invited to attend the talk, and those faculty wishing to remain for the oral exam following are welcome to do so.

1 0

Samyak Gupta will present his General Exam "Attacking Federated Natural Language Modeling and Building Tools for Optimizing Cognitive Models" on Friday, April 15, 2022 at 1:00 PM in CS 302 and via zoom.
by Louis W. Riehl 11 Apr '22

11 Apr '22

Samyak Gupta will present his General Exam "Attacking Federated Natural Language Modeling and Building Tools for Optimizing Cognitive Models" on Friday, April 15, 2022 at 1:00 PM in CS 302 and via zoom. Zoom link: [ https://princeton.zoom.us/my/samyakg | https://princeton.zoom.us/my/samyakg ] Committee Members: Kai Li (advisor), Danqi Chen, Jon Cohen Abstract: In this talk, I will present two distinct areas of my work. In the first half of my talk, I describe an application related to privacy in federated learning. Federated learning allows distributed users to collaboratively train a model while keeping each user’s data private. As a result, it is actively being considered for privacy sensitive applications such as virtual mobile keyboards and analysis of electronic health records in hospitals. A growing body of recent work suggests that a clever eavesdropper may effectively reconstruct image data from gradients transmitted during federated learning. However, little progress has been made regarding the recovery of text data. Moreover, existing literature does not establish any metrics which may be used to quantify the severity of text data leakage. In this presentation, I present the first attack which can reconstruct sentences of private text data during training of a neural language model in federated learning. Our proposed method is built on beam search for text generation, by recovering a set of words from gradients and leveraging the latent memorization of text by large language models. We experiment with a GPT-2 model on several large datasets and demonstrate that our proposed attack can successfully recover private text data during training. Reading List: [ https://docs.google.com/document/d/1t6BU6TflwycaOe7XECeUk7Tln19Aui9JqVOLhxk… | https://docs.google.com/document/d/1t6BU6TflwycaOe7XECeUk7Tln19Aui9JqVOLhxk… ] Everyone is invited to attend the talk, and those faculty wishing to remain for the oral exam following are welcome to do so. Louis Riehl Graduate Administrator Computer Science Department, CS213 Princeton University (609) 258-8014

1 0

UPDATE: Tianyu Gao will present his General Exam "SimCSE: Simple Contrastive Learning of Sentence Embeddings" on Wednesday, April 13, 2022 at 3:00 PM in CS 402 and via Zoom.
by Louis Riehl 11 Apr '22

11 Apr '22

Tianyu Gao will present his General Exam "SimCSE: Simple Contrastive Learning of Sentence Embeddings" on Wednesday, April 13, 2022 at 3:00 PM in CS 402 and via Zoom. Zoom link: https://princeton.zoom.us/j/95387462065 Committee Members: Danqi Chen (advisor), Karthik Narasimhan, Sanjeev Arora Abstract: We present SimCSE, a simple contrastive learning framework that greatly advances the state-of-the-art sentence embeddings. We first describe an unsupervised approach, which takes an input sentence and predicts itself in a contrastive objective, with only standard dropout used as noise. This simple method works surprisingly well, performing on par with previous supervised counterparts. We find that dropout acts as minimal data augmentation and removing it leads to a representation collapse. Then, we propose a supervised approach, which incorporates annotated pairs from natural language inference datasets into our contrastive learning framework, by using "entailment" pairs as positives and "contradiction" pairs as hard negatives. We evaluate SimCSE on standard semantic textual similarity (STS) tasks, and our unsupervised and supervised models using BERT base achieve an average of 76.3% and 81.6% Spearman's correlation respectively, a 4.2% and 2.2% improvement compared to previous best results. We also show -- both theoretically and empirically -- that contrastive learning objective regularizes pre-trained embeddings' anisotropic space to be more uniform, and it better aligns positive pairs when supervised signals are available. Reading List: https://docs.google.com/document/d/1fSGNfvOSkPggmMZUv_rAp4MK0fo05EeL9enjoyep 0Og/edit?usp=sharing Everyone is invited to attend the talk, and those faculty wishing to remain for the oral exam following are welcome to do so.

1 0

Ilya Chugunov will present his General Exam "Implicit Inverse Imaging with Explicit Priors" on Monday, April 18, 2022 at 10 am in Friend Center 202 and via Zoom.
by Louis W. Riehl 11 Apr '22

11 Apr '22

Ilya Chugunov will present his General Exam "Implicit Inverse Imaging with Explicit Priors" on Monday, April 18, 2022 at 10 am in Friend Center 202 and via Zoom. Zoom link: [ https://princeton.zoom.us/j/91472663309 | https://princeton.zoom.us/j/91472663309 ] Committee Members: Felix Heide (advisor), Szymon Rusinkiewicz, Jia Deng Abstract: Armed with the right math, we can solve many tough imaging problems analytically, with no train time and no train data. Unfortunately, from lens aberrations to high level semantic features, the real visual world is full of things we either can’t or don’t want to model, but which we can learn from data. After all, we’ve yet to mathematically define a dog, but even a simple U-Net can be trained to recognize one. In this exam I present three works which try to combine analytic priors with a learned reconstruction pipeline: Mask-ToF, which uses differentiable simulation of time-of-flight imaging to help learn a physical augmentation to depth camera hardware in order to reduce imaging artifacts; GHz-ToF, which uses this same simulation to train a network for correction ambiguity in phase measurements; and my most recent work on handheld neural depth refinement, wherein we train an implicit model to recover fine depth of tabletop objects from a single smartphone camera snapshot. Reading List: [ https://docs.google.com/document/d/1tLs9TnnLzUGX6ebPPVpXsvLgsiCirsKl2Jxo_ji… | https://docs.google.com/document/d/1tLs9TnnLzUGX6ebPPVpXsvLgsiCirsKl2Jxo_ji… ] Everyone is invited to attend the talk, and those faculty wishing to remain for the oral exam following are welcome to do so.

1 0

Thomas Macrina will present his Pre-FPO "Precise Alignment of Serial Section Electron Microscopy Images and Analysis of Neural Circuits" on Tuesday, April 12, 2022 at 5 pm EDT via Zoom.
by Louis W. Riehl 08 Apr '22

08 Apr '22

Thomas Macrina will present his Pre-FPO "Precise Alignment of Serial Section Electron Microscopy Images and Analysis of Neural Circuits" on Tuesday, April 12, 2022 at 5 pm EDT via Zoom. Zoom link: https://princeton.zoom.us/j/99057082150 Committee Members: H. Sebastian Seung(Examiner, advisor), Mala Murthy (Examiner), Felix Heide(examiner), Kai Li (reader), Stephan Saalfeld (HHMI Janelia, reader). All are welcome to attend. Title: Precise Alignment of Serial Section Electron Microscopy Images and Analysis of Neural Circuits Abstract: Neural circuits provide a means to test how the structure of a brain is related to its function, but reconstructing complete circuits in awhile mammalian brain is currently not possible. A promising path is to image a brain with serial section electron microscopy and use automatic methods tore construct circuits from the data. The imaging process introduces significant physical distortion that causes the automatic methods to make errors, so it's crucial to remove the distortion in a step called alignment. There are existing approaches to alignment, and by augmenting them with human-in-the-loop intervention we were able to produce a very accurate alignment of a five teravoxel mouse cortex dataset. This approach failed to correct distortion caused by cracks and folds, and the amount of human effort involved made it prohibitively expensive to process larger datasets. To overcome these limitations, our group developed new alignment pipeline that uses deep optic flow models to predict more precise distortion measurements. Using this new alignment pipeline, we have been able to accurately align much larger datasets, including a 100 teravoxel whole fly brain and a one petavoxel cubic millimeter of mouse cortex. These alignments enabled high quality automatic reconstructions that are now being proofread with manageable human effort to produce some of the largest neural circuits to date. With improvements in the reconstruction technology, like our new alignment pipeline, neural circuits are becoming more accessible, and there will be more effort put toward analyzing them. We analyzed a neural circuit reconstructed from the terascale mouse cortex dataset, finding that sizes of synapses in a shared layer 2/3 pyramidal neuron connection have a correlated binary component, and that the network of these pyramidal neurons is well-described by a configuration model. Both results may have implications for how to constrain biological models of memory or neural network organization. Louis Riehl Graduate Administrator Computer Science Department, CS213 Princeton University (609) 258-8014

1 0

Robotics Seminar: Cameron Aubin TODAY at 10:45am
by Emily C. Lawrence 07 Apr '22

07 Apr '22

ROBOTICS SEMINAR Speaker: Cameron Aubin Title: Towards enduring autonomous robots via embodied energy Day: Thursday, April 7, 2022 Time: 10:45-11:45am Location: Friend Center Convocation Room (FC113) Host: Glaucio Paulino Abstract: Modern robots lack the combination of endurance and adaptability that we see in living organisms. One explanation for this phenomenon is that robots are typically composed of individual power, actuation, sensory, and control blocks, each existing as separate materials and optimized for different tasks. Conversely, biological systems are highly interconnected, hierarchical, and capable of self-assembly, thereby enabling increased complexity and multifunctionality. Many robots also house energy in a single storage unit, often a battery, while more advanced organisms distribute energy throughout their entire bodies. In this talk, I will discuss my work developing robots with multifunctional embodied energy systems that increase their performance and endurance. My discussion will focus on the creation of an untethered soft robotic fish containing a “synthetic vascular system” inspired by redox flow batteries. This system combines the functions of mechanical actuation with hydraulic force transfer and electrolyte-based energy storage, increasing the energy density of the robot by over 300%. Next, I will discuss how energy-dense chemical fuels can be used to create powerful (Pstroke > 50W, Pspec = 147W/g) soft microactuators. We have leveraged microscale combustion and localized flame quenching to develop two demonstrations: a soft haptic display that operates at high frequencies (ƒ > 1kHz) without electromechanical valves and a quadrupedal microrobot capable of multigait/directional movement. Using energy-dense fuels at this scale can increase operating time, power density, and ultimately untether microrobotic systems. Bio: Cameron is a PhD candidate in the Department of Mechanical and Aerospace Engineering at Cornell University. He is a member of the Organic Robotics Laboratory and is advised by Dr. Robert Shepherd. Cameron’s work centers on improving the endurance, adaptability, and autonomy of robots by embodying them with multifunctional, biologically-inspired energy systems. His interests also include soft robotics, advanced manufacturing, and microrobotics. Cameron earned a B.S.E in Biomedical Engineering from Duke University in 2014 and an M.S. in Mechanical Engineering from Cornell University in 2019. His is a recipient of the Harriet Davis Fellowship and his research has been featured in several popular media outlets, including CNN, PBS, BBC, Wired, and more.

1 0

CS Colloquium speaker Natasha Jaques: TODAY at 12:30pm
by Emily C. Lawrence 07 Apr '22

07 Apr '22

CS Colloquium Speaker Speaker: Natasha Jaques, Google Brain / University of California, Berkeley Date: Thursday, April 7, 2022 Time: 12:30pm EST Location: CS 105 Host: Karthik Narasimhan Event page: [ https://www.cs.princeton.edu/events/26182 | https://www.cs.princeton.edu/events/26182 ] This talk will be live-streamed at [ https://mediacentrallive.princeton.edu/ | https://mediacentrallive.princeton.edu/ ] Title: Social Reinforcement Learning Abstract: Social learning helps humans and animals rapidly adapt to new circumstances, coordinate with others, and drives the emergence of complex learned behaviors. What if it could do the same for AI? This talk describes how Social Reinforcement Learning in multi-agent and human-AI interactions can address fundamental issues in AI such as learning and generalization, while improving social abilities like coordination. I propose a unified method for improving coordination and communication based on causal social influence. I then demonstrate that multi-agent training can be a useful tool for improving learning and generalization. I present PAIRED, in which an adversary learns to construct training environments to maximize regret between a pair of learners, leading to the generation of a complex curriculum of environments. Agents trained with PAIRED generalize more than 20x better to unknown test environments. Ultimately, the goal of my research is to create intelligent agents that can assist humans with everyday tasks; this means leveraging social learning to interact effectively with humans. I show that learning from human social and affective cues scales more effectively than learning from manual feedback. However, it depends on accurate recognition of such cues. Therefore I discuss how to dramatically enhance the accuracy of affect detection models using personalized multi-task learning to account for inter-individual variability. Together, this work argues that Social RL is a valuable approach for developing more general, sophisticated, and cooperative AI, which is ultimately better able to serve human needs. Bio: Natasha Jaques holds a joint position as a Senior Research Scientist at Google Brain and Visiting Postdoctoral Scholar at UC Berkeley. Her research focuses on Social Reinforcement Learning in multi-agent and human-AI interactions. Natasha completed her PhD at MIT, where her thesis received the Outstanding PhD Dissertation Award from the Association for the Advancement of Affective Computing. Her work has also received Best Demo at NeurIPS, an honourable mention for Best Paper at ICML, Best of Collection in the IEEE Transactions on Affective Computing, and Best Paper at the NeurIPS workshops on ML for Healthcare and Cooperative AI. She has interned at DeepMind, Google Brain, and was an OpenAI Scholars mentor. Her work has been featured in Science Magazine, Quartz, IEEE Spectrum, MIT Technology Review, Boston Magazine, and on CBC radio. Natasha earned her Masters degree from the University of British Columbia, and undergraduate degrees in Computer Science and Psychology from the University of Regina.

1 0

Tianyu Gao will present her General Exam "SimCSE: Simple Contrastive Learning of Sentence Embeddings" on Wednesday, April 13, 2022 at 3:00 PM via Zoom.
by Louis W. Riehl 06 Apr '22

06 Apr '22

Tianyu Gao will present her General Exam "SimCSE: Simple Contrastive Learning of Sentence Embeddings" on Wednesday, April 13, 2022 at 3:00 PM via Zoom. Zoom link: [ https://princeton.zoom.us/j/95387462065 | https://princeton.zoom.us/j/95387462065 ] Committee Members: Danqi Chen (advisor), Karthik Narasimhan, Sanjeev Arora Abstract: We present SimCSE, a simple contrastive learning framework that greatly advances the state-of-the-art sentence embeddings. We first describe an unsupervised approach, which takes an input sentence and predicts itself in a contrastive objective, with only standard dropout used as noise. This simple method works surprisingly well, performing on par with previous supervised counterparts. We find that dropout acts as minimal data augmentation and removing it leads to a representation collapse. Then, we propose a supervised approach, which incorporates annotated pairs from natural language inference datasets into our contrastive learning framework, by using "entailment" pairs as positives and "contradiction" pairs as hard negatives. We evaluate SimCSE on standard semantic textual similarity (STS) tasks, and our unsupervised and supervised models using BERT base achieve an average of 76.3% and 81.6% Spearman's correlation respectively, a 4.2% and 2.2% improvement compared to previous best results. We also show -- both theoretically and empirically -- that contrastive learning objective regularizes pre-trained embeddings' anisotropic space to be more uniform, and it better aligns positive pairs when supervised signals are available. Reading List: [ https://docs.google.com/document/d/1fSGNfvOSkPggmMZUv_rAp4MK0fo05EeL9enjoye… | https://docs.google.com/document/d/1fSGNfvOSkPggmMZUv_rAp4MK0fo05EeL9enjoye… ] Everyone is invited to attend the talk, and those faculty wishing to remain for the oral exam following are welcome to do so.

1 0

Anat Kleiman will present her MSE talk "Adaptation and Intrinsic Motivations in Reinforcement Learning" on Wednesday, April 13th, 2022 at 2pm via Zoom
by Nicki Mahler 06 Apr '22

06 Apr '22

Anat Kleiman will present her MSE talk "Adaptation and Intrinsic Motivations in Reinforcement Learning" on Wednesday , April 13th, 2022 at 2pm via Zoom. Her committee members are as follows: Ryan Adams (adviser), Karthik Narasimhan (reader) Zoom link: https://princeton.zoom.us/j/92754574624 Please see abstract below. All are welcome to attend. There exist certain discrepancies between how humans and machine learning models learn and make decisions. These discrepancies can not only make it difficult to model human behavior, but can also decrease the efficiency with which models predict and take actions. This thesis will focus on two of these discrepancies, mainly quickly adapting to new tasks using existing knowledge, and the influence of collective, societal goals in individual decision making. It will do this through 2 projects that attempt to provide machine-learning solutions towards allowing for them. The first project focuses on adaptation on the road. By leveraging meta-learning in autonomous vehicle driving, and treating experiences with other drivers as separate tasks, our model learns to generalize knowledge from across different types of drivers. In turn, it is able to quickly adapt to unseen drivers using limited driving experience with them. The second project focuses on collective goal seeking in a multi-agent environment. We use a group intrinsic reward function that is optimized over time to provide agents with an immediate intrinsic reward that they must maximize along with their selfish reward. By using the expected lifetime reward of an agent in the environment to optimize the intrinsic reward function, our agents learn to take possibly sub-optimal actions that benefit the collective group, rather than just themselves.

1 0