- talks - lists.cs.princeton.edu

CS & BioE Seminar Speaker: Ellen Zhong, Monday, Feb 14 at 12:30
by Emily C. Lawrence 14 Feb '22

14 Feb '22

CS & BioE Seminar Speaker: Ellen Zhong, Massachusetts Institute of Technology Date: Monday, Feb 14, 2022 Time: 12:30pm EST Location: CS 105 Host: Clifford Brangwynne, Chemical and Biological Engineering Event page: [ https://www.cs.princeton.edu/events/26161 | https://www.cs.princeton.edu/events/26161 ] *Attendees can view this talk in person or remotely over Zoom here: [ https://princeton.zoom.us/j/98415856885 | https://princeton.zoom.us/j/98415856885 ] **Zoom link is available to Princeton authenticated users only. Title: Machine learning for reconstructing dynamic protein structures from cryo-EM images Abstract: Over the last decade, cryo-electron microscopy (cryo-EM) has emerged as a powerful imaging technology for visualizing the 3D structure of proteins and other biomolecules at near-atomic resolution. Unlike other methods in structural biology, cryo-EM is uniquely poised to image large and dynamic protein complexes. However, this promise is limited by the computational task of 3D reconstruction, where a dataset of noisy and unlabeled 2D projection images are combined to infer the 3D structure(s) of the molecule of interest. In this seminar, I will introduce cryoDRGN, a machine learning system for heterogeneous cryo-EM reconstruction. Underpinning the cryoDRGN method is a deep generative model parameterized by a new neural representation of 3D volumes and a learning algorithm combining exact and variational inference to optimize this model from unlabeled 2D cryo-EM images. Extended to real datasets and released as an open-source tool, cryoDRGN has been used to discover new protein structures and visualize continuous trajectories of their motion. I will discuss various extensions of the method to broaden the scope of cryo-EM to new classes of dynamic protein complexes. Finally, I will discuss how recent advances in machine learning for protein structure prediction (e.g. AlphaFold and Rosettafold) can complement methods for cryo-EM structure determination and what key algorithmic challenges remain to realize the next era of 3D visual biology.

1 2

Zeyu Wang will present his Pre FPO "Tackling Imperfections in Data for Real-world Vision Systems" on Tuesday, February 8, 2022 at 3pm via Zoom
by Nicki Mahler 07 Feb '22

07 Feb '22

Zeyu Wang (ECE) will present his Pre FPO "Tackling Imperfections in Data for Real-world Vision Systems" on Tuesday, February 8, 2022 at 3pm via Zoom Zoom link: https://princeton.zoom.us/j/8491812031 Committee members: Prof. Olga Russakovsky (advisor), Prof. Karthik Narasimhan, Prof. Jia Deng, Prof. Peter Ramadge (ECE). Title: Tackling Imperfections in Data for Real-world Vision Systems Abstract: Data is the fuel of modern neural networks and largely contributes to the rise of deep learning. However, ‘no dataset is perfect’ and various data issues pose significant challenges for building real-world vision systems. In this talk, I will present our works on tackling three prevalent data issues: spurious correlation, long-tail and noise. First, I will introduce a new benchmark for studying how to mitigate bias learned from spurious correlation in data. Then I will talk about a new metric and an inference-time re-ranking technique for tackling long-tail data problems in captioning models. Finally, I will discuss how to utilize multi-query to address data noise for video retrieval systems.

1 0

Yuping Luo will present his Pre FPO "On Designing Efficient and Effective Deep Model-based Reinforcement Learning Algorithms" on Friday, February 4, 2022 at 3pm via Zoom.
by Nicki Mahler 03 Feb '22

03 Feb '22

Yuping Luo will present his Pre FPO "On Designing Efficient and Effective Deep Model-based Reinforcement Learning Algorithms" on Friday, February 4, 2022 at 3pm via Zoom. Zoom link: https://princeton.zoom.us/j/93379363621?pwd=QWFiV0JZT3VwTXVacnc4SEh5UVErdz09 Committee: Sanjeev Arora (advisor, examiner) Karthik Narasimhan (examiner) Tengyu Ma (examiner, Stanford) Elad Hazan (reader) Jason Lee (reader, ECE) All are welcome to attend. Abstract: Tengyu Ma is an outside committee member. He is an Assistant Professor at Stanford University. He graduated from Princeton University, also advised by Prof. Sanjeev Arora (my advisor), I have a lot of collaborations with him and think he can be a great fit for my committee. Abstract: Reinforcement Learning (RL) enables the agent to make decisions in a complex interactive environment. To deal with a large state space, deep reinforcement learning employs neural networks as powerful function approximators. Recent advances in deep reinforcement learning have demonstrated its great potential for real-world problems. However, two concerns prevent RL from being applied: Efficiency and Efficacy. Can we design algorithms to learn a policy with desired property (effectively) with limited information (efficiently)? The property can be high total rewards, safety, or interpretability. To this end, model-based reinforcement learning algorithms use a dynamics model, whether learned or given, to assist policy learning. The access to dynamics models empowers the algorithms to plan, which is key to sequential decision making. In this talk, I will present our works on improving the efficiency and efficacy of reinforcement learning by designing model-based algorithms. I'll first talk about our work on offline reinforcement learning, which can efficiently use expert demonstration. The key idea is to learn a conservatively-extrapolated value function and a dynamics model to make local corrections, so the policy will stay close to the demonstration and achieve high total rewards. Next, I'll talk about our work on effectively learning a policy with safety constraints during training. The key idea is to learn barrier certificates and a dynamics model to certify safe states so that unsafe states can be avoided.

1 0

Hei Law will present his Pre FPO "Learning to Detect Objects by Grouping" on Friday, February 4, 2022 at 2:30pm via Zoom.
by Nicki Mahler 28 Jan '22

28 Jan '22

Hei Law will present his Pre FPO "Learning to Detect Objects by Grouping" on Friday, February 4, 2022 at 2:30pm via Zoom. Zoom link: https://princeton.zoom.us/j/6764372799?pwd=a3U4ZmZBZXdTbWhqOFBHbmN1c2pEUT09 Committee: Prof. Jia Deng (examiner), Prof. Adam Finkelstein (examiner), Prof. Felix Heide (examiner), Prof. Szymon Rusinkiewicz (reader) and Prof. Olga Russakovsky (reader) Abstract: Object detection is a fundamental task in computer vision and a part of many computer vision algorithms. In this talk, we will present our work on developing more accurate and efficient object detectors. We will first present two novel corner-based object detectors: CornerNet and CornerNet-Lite. CornerNet reformulates object detection as detecting pairs of corners, which simplifies the design of modern object detectors, and achieves state-of-the-art results. Following up on CornerNet, we introduce CornerNet-Lite where we explore two ways to improve CornerNet’s efficiency. The first way uses a new attention mechanism to process fewer pixels and achieve better performance, while the second one uses a new compact backbone network to achieve real-time performance. Beside the two new detectors, we also present our ongoing work on designing a new pretext task for object detectors to learn useful visual representations. Inspired by the human vision system, we leverage motion information in videos to determine which pixels are likely to belong to the same objects, and train a network to group pixels by objects before fine-tuning it on the detection task. Everyone is welcome to attend.

1 0

Asiff Hirji lecture rescheduled to Feb 1, 2022
by Emily C. Lawrence 26 Jan '22

26 Jan '22

Speaker: ASIFF HIRJI, President of Figure and Former President and COO of Coinbase Title: DeFi and How Blockchain will Transform Financial Services Day: Tuesday, February 1, 2022 Time: 4:30 pm to 5:30 pm Location: Arthur Lewis Auditorium – Robertson Hall Host: Princeton’s Bendheim Center for Finance, the Julis Rabinowitz Center for Public Policy & Finance, and the Center for Information Technology Policy Open to: Princeton University students, faculty and staff. Event and Registration Page: [ https://bcf.princeton.edu/events/defi-and-how-blockchain-will-transform-fin… | https://bcf.princeton.edu/events/defi-and-how-blockchain-will-transform-fin… ] Abstract: Asiff Hirji, President of Figure Technologies and former President & COO of Coinbase and TD Ameritrade discusses how blockchain, and the DeFi movement, will transform financial services. Through the provision of an immutable ledger, public registry and real-time bilateral transaction settlement capability, the application of blockchain technology will meaningfully reduce the risk, cost and capital requirements associated with current financial systems and infrastructure as well as eliminate the need for most financial intermediaries. The implications for the financial industry will be profound, please join us to learn more. Bio: Asiff Hirji is President of Figure Technologies, Inc. (Figure™), a leader in transforming financial services through the power of blockchain technology. He recently served as President and COO of Coinbase where he grew the company to over $1 billion in revenue; launched new businesses; built out the management team and achieved an $8 billion valuation. Prior to Coinbase, Hirji was an Operating Partner with Andreesen Horowitz focused on helping their best companies scale faster. He previously served as President and COO of TD Ameritrade, growing that business 6x into the nation’s largest online broker. Hirji has also held senior leadership roles with TPG Capital, Saxo Bank, HP, and Bain Capital and has served on a number of public and private boards including Citrix and Advent Software.

1 1

Matthew Myers will present his Pre FPO "Inferring tumor heterogeneity from DNA sequencing data" on Thursday, January 27, 2022 at 3pm via Zoom
by Nicki Mahler 20 Jan '22

20 Jan '22

Matthew Myers will present his Pre FPO "Inferring tumor heterogeneity from DNA sequencing data" on Thursday, January 27, 2022 at 3pm via Zoom. Zoom link: https://princeton.zoom.us/j/94807976020?pwd=d21JY2VLZElXbjlCZXJnbjRwL0svUT09 Committee members: Ben Raphael (adviser), Mona Singh, and Olga Troyanskaya; Readers: Yuri Pritykin and Quaid Morris (Sloan Kettering Institute) All are welcome to attend. Title: Inferring tumor heterogeneity from DNA sequencing data Abstract: Cancer is an evolutionary process where cells acquire somatic mutations over time at various genomic scales, from single-position changes (single-nucleotide variations or SNVs), to changes in the number of copies (copy number aberrations or CNAs) of larger regions of the genome, to duplication of the entire genome (whole-genome duplication or WGD). As a result of this process, each tumor is heterogeneous -- it consists of a mixture of different populations of cells, or clones, each characterized by a distinct set of mutations. Understanding these clones and their evolution is critical to treating many cancers. In this talk, I will present three computational methods for inferring tumor heterogeneity from DNA sequencing data. First, I will present CALDER, which uses SNVs from longitudinal bulk sequencing samples to infer a phylogenetic tree which defines the tumor clones and their evolutionary relationships. Unlike prior methods, CALDER applies constraints derived from the sequential ordering of samples to produce more plausible trees. Next, I will present HATCHet2, which infers CNAs from one or more bulk DNA sequencing samples. HATCHet2 combines several innovations including reference-based phasing and location-aware clustering with the novel factorization approach of HATCHet. Finally, I will present SBMClone, which uses SNVs identified from ultra-low-coverage single-cell DNA sequencing data to group tumor cells. SBMClone uses a stochastic block model to distinguish tumor cells in data that was too sparse for previous methods.

1 0

Zachary Teed will present his Pre FPO "Optimization Inspired Neural Networks for Multiview 3D" on Friday, January 21, 2022 at 4pm via Zoom.
by Nicki Mahler 18 Jan '22

18 Jan '22

Zachary Teed will present his Pre FPO "Optimization Inspired Neural Networks for Multiview 3D" on Friday, January 21, 2022 at 4pm via Zoom. Zoom link: https://princeton.zoom.us/j/2013757951 Committee members: Jia Deng (Adviser) Adam Finkelstein Felix Heide Olga Russakovsky Syzmon Rusinkiewicz All are welcome to attend. Title: Optimization Inspired Neural Networks for Multiview 3D Abstract: Multiview 3D has traditionally been approached as an optimization problem. The solution is produced by an algorithm which searches over continuous variables (camera pose, depth, 3D points) to satisfy both geometric constraints and visual observations. In contrast, deep learning offers an alternative strategy where the solution is produced by a general-purpose network with learned weights. In this talk, I will be discussing a hybrid approach for multiview problems, where we explore neural architecture designs inspired by optimization. I then demonstrate this general strategy in building accurate and robust systems for optical flow, stereo, scene flow, and visual SLAM.

1 0

TigerTalks Digital: Gaming Tigers
by Emily C. Lawrence 17 Jan '22

17 Jan '22

[ https://entrepreneurs.princeton.edu/events/tigertalks-digital-gaming-tigers… | TigerTalks Digital: Gaming Tigers ] Join the Princeton Entrepreneurship Council on Thursday, January 20th, 4:30 PM ET for an engaging conversation with four Princeton alumni founders in the gaming space who will share their stories (and also show brief demonstrations of their latest products). Each of their journeys is remarkable. Harry Hagerty ’82 (CFO of Galaxy Gaming) has been a top executive in the gaming industry for two decades. Galaxy Gaming develops, manufactures and distributes innovative proprietary table games, state-of-the-art electronic wagering platforms and enhanced bonusing systems to land-based, riverboat, cruise ships and online casinos worldwide. He will provide perspectives on the traditional gaming business and how it has evolved to include virtual, online experiences. Ted Price ’90 (Founder, President and CEO of Insomniac Games) who started the company a few years after graduation and established it as one of the industry’s leading game developers over the past three decades. Imagine sustaining success in the gaming industry for close to 30 years, selling the company to Sony Entertainment and continuing to lead it while making ever-bigger games. Ted will also share clips from two of Insomniac’s recent Marvel/PlayStation product launches. Dave Getson ’94 (CEO of gnet) leads one of the gaming industry’s leading marketing firms. LA-based gnet is a strategic and creative agency that has helped launch and grow some of the biggest gaming brands of the past two decades for partners including Microsoft Xbox, Activision Blizzard, Bethesda Softworks and many more. David brings perspective on the unique relationship between successful games and their highly engaged fan communities. Derek Chan ’10 (Founder and CEO of Appslots) represents the newest breed of gaming executive, focusing on the hot crypto, NFT and metaverse space. Imagine starting career as Wall Street trader, then saw an opportunity in mobile games and eventually won the official Google Play Top Developer recognition for a cumulative work of 25+ top grossing categories mobile games and 10 million+ users. Is metaverse the new opportunity? How can we capture it? Derek will provide unique perspectives on the industry’s newest buzzwords, discuss the potential benefits and challenges, and walkthrough blockchain games’ underlying concepts with one of his latest projects as an example. The panel of Gaming Tigers will talk about the gaming industry, discuss their creative accomplishments, and answer your questions. Register here: [ https://entrepreneurs.princeton.edu/events/tigertalks-digital-gaming-tigers | https://entrepreneurs.princeton.edu/events/tigertalks-digital-gaming-tigers ]

1 0

Sunnie Kim will present her General Exam "Evaluating the Human Interpretability of Visual Explanations" on Thursday, January 20, 2022 at 10am via Zoom.
by Nicki Mahler 13 Jan '22

13 Jan '22

Sunnie Kim will present her General Exam "Evaluating the Human Interpretability of Visual Explanations" on Thursday, January 20, 2022 at 10am via Zoom. Zoom link: [ http://princeton.zoom.us/my/sunniesuhyoung | http://princeton.zoom.us/my/sunniesuhyoung ] Committee members: Olga Russakovsky (adviser) Andrés Monroy-Hernández Ruth Fong Abstract: As machine learning is increasingly applied to high-impact, high-risk domains, there have been a number of new methods aimed at making machine learning models more human interpretable. Despite the recent growth of interpretability work, there is a lack of systematic evaluation of proposed techniques. In this exam, I will present my recent work, HIVE: Evaluating the Human Interpretability of Visual Explanations, where I propose HIVE (Human Interpretability of Visual Explanations), a novel human evaluation framework for interpretability methods in computer vision. While human studies should be the gold standard in properly evaluating how interpretable a method is to human users, they are often avoided due to challenges associated with cost, study design, and cross-method comparison. I will discuss how HIVE mitigates these issues and enables the evaluation of diverse visual explanations. To demonstrate the extensibility and applicability of HIVE, I conducted IRB-approved studies of four existing interpretability works: GradCAM, BagNet, ProtoPNet, and ProtoTree. Results suggest that explanations (regardless of if they are actually correct) engender human trust, yet are not distinct enough for users to distinguish between correct and incorrect predictions. I will conclude the talk with a discussion of key insights and propositions for future research. Reading List: [ https://docs.google.com/document/d/1WL1yhvTS1yFwKd3E7C55AIMjCAdmBXsAsNVWHb9… | https://docs.google.com/document/d/1WL1yhvTS1yFwKd3E7C55AIMjCAdmBXsAsNVWHb9… ] 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

Joshua Cohen will present his General Exam "Verified Forward Erasure Correction with Coq and VST" on Thursday, January 13, 2022 at 1pm via Zoom.
by Nicki Mahler 06 Jan '22

06 Jan '22

Joshua Cohen will present his General Exam "Verified Forward Erasure Correction with Coq and VST" on Thursday, January 13, 2022 at 1pm via Zoom. Zoom link: https://princeton.zoom.us/j/92180386467 Committee members: Andrew Appel (Adviser), Aarti Gupta, and Kyle Jamieson Abstract: Most methods of data transmission and storage are prone to errors, leading to data loss. Error-correcting codes encode the data with redundant parity information to allow the original data to be recovered even in the presence of a small number of errors. An erasure code uses an ECC to correct erasures, errors where the location is known. There are dozens of classes of error-correcting and erasure codes, many based on sophisticated mathematics, making them difficult to verify using automated tools. These erasure codes can be implemented in networks as Forward Erasure Correction, in which a sender encodes a block of input packets using an ECC, and the receiver decodes if errors were detected. In this talk, I will describe my verification of a real-world C implementation of FEC using Coq and the Verified Software Toolchain for verifying C programs. This verification includes both a purely-mathematical proof that the underlying algorithm (a form of Reed-Solomon erasure coding) is correct and the VST proof that the C code, over 25 years old and including many modifications and optimizations, correctly implements this algorithm. Reading list: [ https://docs.google.com/document/d/19_AqGyWpsvNnEOHIlYKJhv6D4GXR9JX6ezQi6Bf… | https://docs.google.com/document/d/19_AqGyWpsvNnEOHIlYKJhv6D4GXR9JX6ezQi6Bf… ] 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