- talks - lists.cs.princeton.edu

New Seminar Series - ML & Physics
by Elizabeth Wang 06 Sep '23

06 Sep '23

Greetings, Professor Boris Hanin from Operations Research & Financial Engineering would like to share the following seminar series with COS students, postdocs, and faculty. If you have any questions, please contact Boris at [ mailto:bhanin@princeton.edu | bhanin(a)princeton.edu ] . >From Boris: Dear Colleagues, We wanted to bring to your attention a new seminar series, Machine Learning in Physics, focused on interdisciplinary work at the intersection of physics and ML. Here are the details: What ? A seminar series highlighting research on both physics-inspired approaches to understanding ML and the use of ML for physics applications When/Where ? Wednesdays 4:30 - 5:30pm in Jadwin Hall, Room A10 with light refreshments after the presentation. Who ? The speakers for Fall 2023 are as follows: * Sept. 27 : Shirley Ho (Flatiron) * Oct. 4 : Juan Felipe Carrasquilla Alvarez (Vector) * Nov. 1: Javier Duarte (UCSD) * Nov. 15 : Phiala Shananan (MIT) * Nov 29 : Andy Leifer (Princeton) For more details — and to sign up for the seminar mailing list — please see the seminar [ https://ml.physics.princeton.edu/ | website ] . Best, Boris Hanin, Gage DeZoort, Dan Marlow, Mariangela Lisanti, Peter Elmer (seminar co-organizers)

1 0

“Tony” Runzhe Yang will present his FPO "From mind to machine: neural circuits, learning algorithms, and beyond." on Tuesday, September 26, 2023 at 12:00 PM in CS 105 and Zoom.
by CS Grad Department 06 Sep '23

06 Sep '23

“Tony” Runzhe Yang will present his FPO "From mind to machine: neural circuits, learning algorithms, and beyond." on Tuesday, September 26, 2023 at 12:00 PM in CS 105 and Zoom. Location: Zoom Link: [ https://princeton.zoom.us/j/91084214303 | https://princeton.zoom.us/j/91084214303 ] The members of Runzhe’s committee are as follows: Examiners: H. Sebastian Seung (Adviser), Karthik Narasimhan (Adviser), Mala Murthy Readers: Ryan Adams, Dmitri Chklovskii (Flatiron Institute; NYU) A copy of his thesis is available upon request. Please email gradinfo(a)cs.princeton.edu if you would like a copy of the thesis. Everyone is invited to attend his talk. Abstract follows below: This thesis explores diverse topics within computational neuroscience and machine learning. The work begins by examining the organization of biological neuronal circuits re- constructed by electron microscopy. First, our study of neural connectivity patterns in the mouse primary visual cortex illustrates the necessity for refined understanding of the non- random features of cortical connections, challenging conventional perspectives. Second, in the larval zebrafish hindbrain, our novel discovery highlights overrepresented three-cycles of neuron, an observation unprecedented in electron microscopy-reconstructed neuronal wiring diagrams. Additionally, I present an exhaustive compilation of motif statistics and network characteristics for the complete adult Drosophila brain. These efforts collectively enrich our understanding of the intricate wiring diagram of neurons, offering new insights into the organizational principles of biological brains. In the second part of the thesis, I introduce three distinct machine learning algorithms. The first algorithm, a biologically plausible unsupervised learning algorithm, is implemented within artificial neural networks using Hebbian feedforward and anti-Hebbian lateral connections. The theoretical discourse explores the duality and convergence of the learning process, connecting with the generalized concept of the “correlation game” principle. The second algorithm presents a novel multi-objective reinforcement learning approach, adept at managing real-world scenarios where multiple potentially conflicting criteria must be optimized without predefined importance weighting. This innovation allows the trained neural network model to generate policies that align optimally with user- specified preferences across the entire space of preference. The third algorithm employs a cognitive science-inspired learning principle for dialog systems. The designed system engages in negotiation with others, skillfully inferring the intent of the other party and predicting how its responses may influence the opponent’s mental state. Collectively, these contributions shed light on the complexities of neural circuit organization and offer new methodologies in machine learning. By examining intelligence from both biological and computational perspectives, the thesis presents insights and reference points for future research, contributing to our growing understanding of intelligence.

1 0

CS seminar speaker Sriram Sankararaman: Tuesday, September 5 at 4:30pm
by Emily C. Lawrence 29 Aug '23

29 Aug '23

Speaker: Sriram Sankararaman, from UCLA Date: Tuesday, September 5, 2023 Time: 4:30pm ET Location: CS 105 Host: Ben Raphael Event page: [ https://www.cs.princeton.edu/events/26490 | https://www.cs.princeton.edu/events/26490 ] Title: The Role of Archaic Admixture in Human Evolution Abstract: Over the past decade, the ability to sequence genomes from both present-day and archaic humans (including our closest evolutionary relatives, the Neanderthals) has transformed our understanding of human history. Analyzing these genome sequences paints a picture of human history in which present-day humans migrated out of Africa but exchanged genes with multiple archaic human populations. I will describe statistical methods that identify segments of DNA inherited from archaic humans that are surviving in our genomes today and how these maps of introgressed archaic DNA are providing insights into human migration and biology. Despite this progress, our understanding of the contribution of archaic introgression to populations in Africa remains limited, in part due to the challenges in obtaining ancient DNA in Africa. Leveraging recently developed approaches that enable inferences about archaic populations without access to their genome sequences, we show that west African populations today inherit substantial genetic ancestry from an as-yet-unidentified archaic ghost population that diverged prior to the split of modern humans and Neanderthals. Finally, we combine maps of introgressed Neanderthal DNA with phenotypic datasets collected in hundreds of thousands of individuals to assess the contribution of introgressed Neanderthal DNA to complex traits. I will discuss the implications of these results for our understanding of human evolution as well as the statistical challenges that need to be solved in this endeavor. Bio: Sriram Sankararaman is a professor in the Departments of Computer Science, Human Genetics, and Computational Medicine at UCLA. His research interests lie at the interface of computer science, statistics and biology. His lab develops machine learning algorithms to analyze genomic data and biomedical data with the broad goal of understanding the interplay between evolution, genomes and traits. He received a B.Tech. in Computer Science from the Indian Institute of Technology, Madras, a Ph.D. in Computer Science from UC Berkeley and was a postdoctoral fellow in Harvard Medical School before joining UCLA. He is a recipient of a NSF Career Award, NIH Pathway to Independence Award, and fellowships from Microsoft Research, the Sloan Foundation, the Okawa Foundation and the Simons Institute.

1 0

Princeton Symposium on Biological & Artificial Intelligence - October 19-20, 2023 - Free Registration
by Elizabeth Wang 17 Aug '23

17 Aug '23

Please see attached and below regarding an upcoming symposium.

1 0

Mingzhe Wang will present his FPO "Deep Learning for Automated Theorem Proving" on Monday, August 21, 2023 at 3:00 PM in CS 301 and Zoom.
by CS Grad Department 07 Aug '23

07 Aug '23

Mingzhe Wang will present his FPO "Deep Learning for Automated Theorem Proving" on Monday, August 21, 2023 at 3:00 PM in CS 301 and Zoom. Location: Zoom link: [ https://princeton.zoom.us/j/3294749300 | https://princeton.zoom.us/j/3294749300 ] The members of Mingzhe’s committee are as follows: Examiners: Jia Deng (Adviser), Danqi Chen, Karthik Narasimhan Readers: Andrew Appel, Sanjeev Arora A copy of his thesis is available upon request. Please email gradinfo(a)cs.princeton.edu if you would like a copy of the thesis. Everyone is invited to attend his talk. Abstract follows below: Automated theorem proving is a fundamental AI task with broad applications to the formalization of mathematics, software and hardware verification, and program synthesis. Deep learning has emerged as a promising approach for guiding theorem provers. In this dissertation, we present our work on developing deep learning approaches for automated theorem proving. First, we propose FormulaNet, a deep learning-based approach to the problem of premise selection. FormulaNet represents a logical formula as a graph that is invariant to variable renaming and then embeds the graph into a vector via a novel graph embedding method that preserves the information of edge ordering. Our approach achieves state-of-the-art results on the HolStep dataset, improving the classification accuracy from 83% to 90.3%. Next, we propose MetaGen, a neural generator that automatically synthesizes theorems and proofs for the purpose of training a theorem prover. On real-world tasks, we demonstrate that synthetic data from our approach improves the theorem prover and advances the state of the art of automated theorem proving in Metamath. Finally, we extend our theorem generator to the interactive theorem prover Lean. We propose TermGen, a neural generator that automatically synthesizes theorems and proofs in Lean by directly constructing proof terms. We also propose a method of expert iteration for training TermGen to synthesize short theorems. At last, we present a case study of generating formal specifications of while-loop programs through a rule-based theorem generator.

1 0

Sven Dorkenwald will present his FPO "Analysis of neuronal wiring diagrams" on Friday, August 18, 2023 at 1:00 PM in Neuroscience Institute A32 and Zoom.
by CS Grad Department 01 Aug '23

01 Aug '23

Sven Dorkenwald will present his FPO "Analysis of neuronal wiring diagrams" on Friday, August 18, 2023 at 1:00 PM in Neuroscience Institute A32 and Zoom. Location: Zoom link: [ https://princeton.zoom.us/j/97076372854 | https://princeton.zoom.us/j/97076372854 ] The members of Sven’s committee are as follows: Examiners: H. Sebastian Seung (Adviser), Michael Freedman, Mala Murthy Readers: Felix Heid, Viren Jain (Google) A copy of his thesis is available upon request. Please email gradinfo(a)cs.princeton.edu if you would like a copy of the thesis. Everyone is invited to attend his talk. Abstract follows below: To understand how the brain works, one must know its neurons and their connections. Maps of synaptic connections between neurons can be created by acquiring and analyzing electron microscopic (EM) brain images, but until recently, even the largest datasets were insufficient to map full neuronal circuits in most organisms or even small brains. Advances in automated EM image acquisition and analysis have now given rise to neuronal reconstructions of mammalian brain circuits and entire insect brains. Errors in these automated reconstructions must still be corrected through proofreading. This thesis introduces new methods to bridge the gap between automated neuron reconstructions and the analysis of neuronal wiring diagrams. First, it presents a proofreading infrastructure that facilitates correction of whole neurons in up to petascale (∼ 1mm3 brain tissue) datasets by communities of scientists and proofreaders. Second, it embeds this proofreading system into an analysis infrastructure to enable queries of the connectome data while it is actively being edited. Third, it presents a self-supervised learning technique for efficient inference of semantic information which is crucial for circuit analyses. These methods have already been used by hundreds of users and enabled numerous neuroscientific analyses. Additionally, this thesis presents the analysis of the largest connectivity map to date between cortical neurons of a defined type (L2/3 pyramidal neurons in mouse visual cortex) which identified constraints on the learning algorithms employed by the cortex. Finally, this thesis reports on the completion of the wiring diagram of the Drosophila melanogaster brain containing ∼130,000 neurons and describes how this work enabled it. The Drosophila connectome was created through a years-long community-based effort and is an incredible resource for the science community and a milestone for connectomics on the way to large mammalian brains. The methods presented in this thesis will be useful for the analysis of larger brain samples as we aim for connectomes of whole mammalian brains.

1 0

Sulin Liu (ECE) will present "Scalable and Interpretable Learning with Probabilistic Models for Knowledge Discovery" on Aug 7, 2023 at 2:30 pm in CSML Room 103 & Zoom
by Gradinfo 01 Aug '23

01 Aug '23

Sulin Liu (ECE) will present "Scalable and Interpretable Learning with Probabilistic Models for Knowledge Discovery" on Aug 7, 2023 at 2:30 pm in CSML Room 103 & Zoom Details can be found here: [ https://ece.princeton.edu/events/scalable-and-interpretable-learning-probab… | https://ece.princeton.edu/events/scalable-and-interpretable-learning-probab… ] Abstract: Novel machine learning methods are at the heart of a transformation underway in science and engineering. Probabilistic models have served as the foundation learning models for knowledge discovery. As surrogate models, they enable efficient black-box optimization or active learning of the system behavior under limited budget to evaluate/query the complex system. Another important use case is to use probabilistic models as generative models to generate de novo designs with desired properties or samples from the equilibrium distribution of physical systems. However, to fully unleash the potential of probabilistic models for knowledge discovery, it is imperative to develop models that are scalable to growing data size and complexity while remaining interpretable to domain experts. In this dissertation, I start with the development of a novel approach that addresses the sparse solution identification problem in Bayesian optimization with probabilistic surrogate models. The discovery of sparse solutions not only enhances the interpretability of the solutions to humans for understanding the system behavior but also facilitates easier streamlined deployment and maintenance with a reduced number of parameters. Next, I present a novel approach utilizing deep learning to enhance the scalability of Gaussian process inference. Gaussian processes are extensively used as probabilistic surrogate models in knowledge discovery, but their practical usage is hindered by the high cost associated with the identification of kernel hyperparameters in GP regression, which involves costly marginal likelihoods. I show how to side-step the need for expensive marginal likelihoods by employing “amortized” inference over hyperparameters. This is achieved through training a single neural network, which consumes a set of data and produces an estimate of the kernel function, useful across different tasks. Finally, I introduce a new family of generative models — marginalization models — for high-dimensional discrete data, which is ubiquitous in science discovery. Marginalization models offer scalable and flexible generative modeling with tractable likelihoods through explicit modeling of all induced marginal distributions with neural networks. Direct modeling of the marginals enables efficient marginal inference and scalable training of any-order generative models for sampling from a given (unnormalized) probability function, which overcomes major limitations of previous methods with tractable likelihoods Zoom: [ https://princeton.zoom.us/j/95051121688?pwd=MFlkRUN6YW5VNmtqMkREdklZSTMxZz09 | https://princeton.zoom.us/j/95051121688?pwd=MFlkRUN6YW5VNmtqMkREdklZSTMxZz09 ] Advisers: Peter Ramadge & Ryan Adams

1 0

Aquinas Hobor (University College London) July 28 at 10:30
by Michael Estepp 25 Jul '23

25 Jul '23

Friday, July 28, 10:30am in room CS105: Programming Languages Seminar Title: Smart Learning to Find Dumb Contracts (by Abdelaziz & Hobor, to appear in USENIX Security 2023) Speaker: Aquinas Hobor (University College London) Abstract: We introduce the Deep Learning Vulnerability Analyzer (DLVA), a vulnerability detection tool for Ethereum smart contracts based on powerful deep learning techniques for sequential data adapted for bytecode. We train DLVA to judge bytecode even though the supervising oracle, Slither, can only judge source code. DLVA's training algorithm is general: we "extend" a source code analysis to bytecode without any manual feature engineering, predefined patterns, or expert rules. DLVA's training algorithm is also robust: it overcame a 1.25% error rate mislabelled contracts, and-the student surpassing the teacher-found vulnerable contracts that Slither mislabelled. In addition to extending a source code analyser to bytecode, DLVA is much faster than conventional tools for smart contract vulnerability detection based on formal methods: DLVA checks contracts for 29 vulnerabilities in 0.2 seconds, a 10-1,000x speedup compared to traditional tools. We benchmark DLVA against nine well-known smart contract analysis tools. Despite using much less analysis time, DLVA completed every query, leading the pack with an average accuracy of 99.7%, pleasingly balancing high true positive rates with low false positive rates.

1 0

Qinshi Wang will present his FPO "Foundationally Verified Data Plane Programming" on Thursday, August 24, 2023 at 2:30pm in CS 402
by Nicki Mahler 25 Jul '23

25 Jul '23

Qinshi Wang will present his FPO "Foundationally Verified Data Plane Programming" on Thursday, August 24, 2023 at 2:30pm in CS 402. The members of his committee are as follows: Readers: David Walker, Nate Foster (Cornell) Examiners: Andrew Appel (Adviser), Jennifer Rexford, Zachary Kincaid All are welcome to attend. Abstract: P4 is a major standardized programming language for programming and specifying the network data plane. P4 is widely applied in a variety of network functionalities, including monitoring, traffic management, forwarding, and security. Recently, stateful applications have been emerging in this area, as supported by programmable hardware. Typical stateful applications include network telemetry (heavy hitters, distributed denial-of-service (DDoS) detection, performance monitoring), middleboxes (firewalls, network address translation (NAT), load balancers, intrusion detection), and distributed services (in-network caching, lock management, conflict detection). Their complexity and rich properties are beyond the ability of existing P4 verifiers. In this thesis, we propose Verifiable P4: a new framework for P4 program verification based on interactive theorem proving that is (1) capable of proving multi-packet properties, (2) modular in terms of P4 program modules, and (3) foundationally sound with respect to a mechanized formal semantics of P4. In order to achieve these goals, we built (1) a mechanized formal semantics of P4 more comprehensive and convenient than existing formal semantics, (2) a set of program logic rules that are proven sound, and (3) an interactive verification system based on the program logic and Coq tactic mechanism. We verified a stateful firewall fully implemented in P4 using the sliding-window Bloom filter with Verifiable P4 and evaluated its utility.

1 0

Zhenyu Song will present his FPO "Optimizing Content Distribution Network Caches with Machine Learning" on Tuesday, August 1, 2023 at 1pm in CS 402.
by Gradinfo 19 Jul '23

19 Jul '23

Zhenyu Song will present his FPO "Optimizing Content Distribution Network Caches with Machine Learning" on Tuesday, August 1, 2023 at 1pm in CS 402. The members of his committee are as follows: Readers: Ravi Netravali Daniel S. Berger (Microsoft & UW) Examiners: Kai Li (Adviser), Wyatt Lloyd (Adviser), and Ryan Adams All are welcome to attend. Abstract: Content Distribution Networks (CDNs) play a pivotal role in Internet traffic. A key part of this caching mechanism is the eviction algorithm that handles the replacement of old cached objects. The effectiveness of the eviction algorithm significantly influences CDN performance. This dissertation explores the application of machine learning (ML) to optimize cache eviction algorithms in CDNs. The central questions addressed in this work are: how to utilize ML to devise an eviction algorithm that surpasses existing heuristics on byte miss ratio, and how to mitigate the CPU overhead while enhancing the robustness of a learned cache in large-scale deployment. Two major challenges faced in the design of a learning-based cache eviction algorithm include heterogeneous user access patterns across different locations and times, and computational and space overheads. To address these challenges, we developed two ML-based eviction algorithms, Learning Relaxed Belady (LRB) and Heuristic Aided Learned Preference (HALP). LRB is the first CDN cache algorithm to directly approximate the Belady MIN (oracle) algorithm by learning access patterns, providing a significant improvement over traditional eviction algorithms. It demonstrated a WAN traffic reduction of 4-25% across six production CDN traces in our simulation. HALP, on the other hand, achieves low CPU overhead and robust DRAM byte miss ratio improvement by augmenting a heuristic policy with ML. It has shown to reduce DRAM byte miss during peak by an average of 9.1%, with a modest CPU overhead of 1.8%, while deployed in YouTube CDN production clusters.

1 0