January 2024 - Colo - huji-cs listserver

CS Colloquium (05/02): Leo Joskowicz (Hebrew University)
by Sagie Benaim 05 Feb '24

05 Feb '24

Dear all, Next week, we have the pleasure of having Prof. Leo Joskowicz give a talk in the colloquium. The seminar will be held on Monday, February 5th, at 14:00. Location: C220. The title, abstract and bio appear below. Looking forward to seeing you, Sagie and Liat *Title:* Three is better than two and better than one: simultaneous deep learning for lesion changes analysis in medical images. *Abstract:* Monitoring cancer patients undergoing treatment requires the analysis by an expert clinician of radiological images acquired every few months. With improvements in treatments leading to longer patient lifespans and increased accessibility of scanners, longitudinal patient studies have increased in their number of CT and MRI scans, making their interpretation more challenging and time-consuming. While numerous AI and deep learning based methods have been developed for medical image analysis, they fail short of providing a viable solution for radiological follow-up. In this talk, we present a novel fully automatic end-to-end pipeline for the comprehensive detection and segmentation of cancer lesions and the analysis of their evolution over time. The key novelties are: 1) SimU-Net, a simultaneous multi-channel 3D deep learning model trained on pairs of registered scans of each patient that identifies the lesions and their changes based on the lesion and healthy tissue appearance differences; 2) a model-based bipartite graph lesions matching method for the analysis of lesion changes at the lesion level; 3) a method for longitudinal analysis of consecutive scans of a patient based on SimU-Net. We demonstrate the method with experimental studies on liver and lung metastases in CT scans and brain metastases in MRI scans that establish state-of the-art results. *Bio*: Leo Joskowicz has been a Professor at the School of Computer Science and Engineering at the Hebrew University of Jerusalem, Israel since 1995. He is the founder and director of the Computer-Aided Surgery and Medical Image Processing Laboratory (CASMIP Lab). Prof. Joskowicz is a Fellow of the IEEE, ASME, and MICCAI (Medical Image Processing and Computer Aided Intervention) Societies. He is the past President of the MICCAI Society, was the Secretary General of the International Society of Computer Aided Orthopaedic Surgery (CAOS International) and of the International Society for Computer Assisted Surgery (ISCAS). He is the recipient of the 2010 Maurice E. Muller Award for Excellence in Computer Assisted Surgery by CAOS International and the 2007 Kaye Innovation Award. He has published two books and over 300 technical works including conference and journal papers, book chapters, and editorials and has 14 issued patents. He is on the Editorial Boards of Medical Image Analysis, Int. J. of Computer Aided Surgery, and Computer Aided Surgery and has served on numerous related program committees.

1 1

CS Colloquium (29/01): Gal Vardi (TTI-Chicago and Hebrew University)
by Sagie Benaim 29 Jan '24

29 Jan '24

Dear all, Next week, we have the pleasure of having Dr Gal Vardi give a talk in the colloquium. The seminar will be held on Monday, January 29th, at 14:00. Location: C220. The title, abstract and bio appear below. Looking forward to seeing you, Sagie and Liat *Title:* On Implicit Bias and Benign Overfitting in Neural Networks *Abstract:* When training large neural networks, there are typically many solutions that perfectly fit the training data. Nevertheless, gradient-based methods often have a tendency to reach those which generalize well, namely, perform well also on test data. Thus, the training algorithm seems to be implicitly biased towards certain networks, which exhibit good generalization performance. Understanding this “implicit bias” has been a subject of extensive research recently. Moreover, in contradiction to conventional wisdom in machine learning theory, trained networks often generalize well even when perfectly fitting noisy training data (i.e., data with label noise), a phenomenon called “benign overfitting”. In this talk, I will discuss the above phenomena. In the first part of the talk, I will discuss the implicit bias and its implications. I will show how the implicit bias can lead to good generalization performance, but can also have negative implications in the context of susceptibility to adversarial examples and privacy attacks. In the second part of the talk, I will explore benign overfitting and the settings in which it occurs in neural networks. *Bio*: Gal is a postdoctoral researcher at TTI-Chicago and the Hebrew University, hosted by Nati Srebro and Amit Daniely as part of the NSF/Simons Collaboration on the Theoretical Foundations of Deep Learning. Prior to that, he was a postdoc at the Weizmann Institute, hosted by Ohad Shamir, and a PhD student at the Hebrew University, advised by Orna Kupferman. His research focuses on theoretical machine learning, with an emphasis on deep-learning theory.

1 1

CS Colloquium (24/01): Yaniv David (Colombia University)
by Sagie Benaim 24 Jan '24

24 Jan '24

Dear all, Next week, we have the pleasure of having Dr Yaniv David give a talk in the colloquium. The seminar will be held on Wednesday, January 24th at 14:00. Location: C220. The title, abstract and bio appear below. Looking forward to seeing you, Sagie and Liat *Title:* Challenges and Opportunities In Securing Software Supply Chains. *Abstract:* Racing to be first to market and deploy new features, developers rely on many external libraries to underpin their software. Each library uses more libraries, creating vast networks of dependencies that the developers know little about and have no control over, forming a knowledge gap that quickly turns into technical debt. Repaying this debt is difficult, as analyzing or examining all libraries is infeasible, and worse, the debt keeps growing due to frequent library updates. Attackers move quickly to collect on this debt by reverse-engineering security updates into 1-day attacks or injecting malicious code into libraries and data used by the applications. In this talk, I will present the systems I built to tackle these challenges: (1) detecting vulnerable libraries in firmware by comparing multiple significant code segments aligned via re-optimizing and normalizing; (2) streamlining software dependency updates via a production-ready hybrid static-dynamic approach for studying the risks of the update before applying it; (3) detecting rogue updates via trust-domain-based tracking for data-flows between different packages in JavaScript code; and (4) hardening applications against data deserialization attacks via a novel type inference technique we call Static Duck Typing, which is based on object behaviors and usage. *Bio*: Yaniv is a postdoc at Columbia University working with Junfeng Yang. His research focuses on improving the reliability and safety of software. He is broadly interested in securing systems and program analysis. He received his PhD from the Technion in 2020, where he was advised by Eran Yahav.

1 1

CS Colloquium (22/01): Tomer Galanti (MIT) - In Zoom
by Sagie Benaim 22 Jan '24

22 Jan '24

Dear all, Next week, we have the pleasure of having Dr. Tomer Galanti give a talk in the colloquium. The seminar will be held on Monday, January 22nd at 14:00. The seminar will be given virtually in Zoom using the following details: https://huji.zoom.us/j/84520950608?pwd=N1VJVHdsRDVLS3crK2R6UlpwNHR5UT09 Meeting ID: 845 2095 0608 Passcode: 393721 The title, abstract, and bio appear below. Looking forward to seeing you all, Sagie and Liat *Title:* Fundamental Problems in AI: Transferability, Compressibility and Generalization *Abstract:* In this talk, we delve into several fundamental questions in deep learning. We start by addressing the question, "What are good representations of data?" Recent studies have shown that the representations learned by a single classifier over multiple classes can be easily adapted to new classes with very few samples. We offer a compelling explanation for this behavior by drawing a relationship between transferability and an emergent property known as neural collapse. Additionally, we explore why certain architectures, such as convolutional networks, outperform fully-connected networks, providing theoretical support for how their inherent sparsity aids learning with fewer samples. Lastly, I present recent findings on how training hyperparameters implicitly control the ranks of weight matrices, consequently affecting the model's compressibility and the dimensionality of the learned features. Additionally, I will describe how this research integrates into a broader research program where I aim to develop realistic models of contemporary learning settings to guide practices in deep learning and artificial intelligence. Utilizing both theory and experiments, I study fundamental questions in the field of deep learning, including why certain architectural choices improve performance or convergence rates, when transfer learning and self-supervised learning work, and what kinds of data representations are learned with Stochastic Gradient Descent. *Bio*: Tomer Galanti is a Postdoctoral Associate at the Center for Brains, Minds, and Machines at MIT, where he focuses on the theoretical and algorithmic aspects of deep learning. He received his Ph.D. in Computer Science from Tel Aviv University and served as a Research Scientist Intern at Google DeepMind's Foundations team during his doctoral studies. He has published numerous papers in top-tier conferences and journals, including NeurIPS, ICML, ICLR, and JMLR. His work, titled "On the Modularity of Hypernetworks," was awarded an oral presentation at NeurIPS 2020.

1 1

Speaker series
by Sagie Benaim 15 Jan '24

15 Jan '24

Hi All, Please refer to the following information about a speaker series that may be of interest to some of you. Best, Sagie *********************************** At Georgia tech this semester, we in the department of computer science are hosting a Turing Award winner Speaker series where we will have 5 Turing award winners and many other well-known computer scientists and academic leaders giving presentations and discussing with us students their experiences through their careers via Microsoft teams video . We will have many other top computer science departments joining us including the departments, department chairs, and vice chancellors of Tel Aviv University, MIT, Yale, Columbia, UCLA, Northwestern, Vanderbilt, and Bowdoin. This is the website for our events: https://www.turing.rsvp/

1 0

CS Colloquium (15/01): Joachim Matthias Neu (Stanford)
by Sagie Benaim 15 Jan '24

15 Jan '24

Dear all, Next week, we have the pleasure of having Joachim Matthias Neu give a talk in the colloquium. The seminar will be held on Monday, January 15th at 14:00. Location: C220. The title, abstract and bio appear below. Looking forward to seeing you, Sagie and Liat *Title:* Internet-Scale Consensus in the Blockchain Era. *Abstract:* Blockchains have ignited interest in Internet-scale consensus as a vital building block for decentralized applications and services that promise egalitarian access and robustness to faults and abuse. While the study of consensus has a 40+ year tradition, the new Internet-scale setting requires a fundamental rethinking of models, desiderata, and protocols. An emergent key challenge is to simultaneously serve clients with different requirements regarding the two fundamental aspects liveness ("good things happen") and safety ("bad things don't happen"). For different instances of this theme, I present the first protocols that allow optimal liveness-safety tradeoff. Results from this line of work have found adoption in the Ethereum blockchain that powers an ecosystem worth $500bn+. *Bio*: Joachim Neu is a PhD candidate at Stanford University advised by David Tse. His research focuses on Internet-scale consensus as a key enabler for decentralized systems, and spans distributed systems, probabilistic systems analysis, applied cryptography, and networking/communications. Website: https://www.jneu.net/

1 1

CS Colloquium (11/01): Eyal Ofek (Note unusual date and time)
by Sagie Benaim 11 Jan '24

11 Jan '24

Dear all, This week, we have the pleasure of having Dr. Eyal Ofek give a talk in the colloquium. The seminar will be held on Thursday, January 11th at 13:00. Location: C220. The title, abstract and bio appear below. Looking forward to seeing you, Sagie and Liat *Title:* Context adaptive applications. *Abstract:* As more people work from home or during travel, new opportunities and challenges arise around mobile office work. They may need to work in makeshift spaces with less than optimal working conditions; applications are not flexible to their physical and social context, and remote collaboration does not account for the difference between the user's conditions and capabilities. Using scene understanding, the ability to augment users' senses to disconnect them from physical limitations, and new architecture for developing applications that share the space with users can change how information workers work. I will display my research in various opportunities for realizing a mobile immersive working environment. *Bio*: Eyal Ofek, Ph.D. I received my Ph.D. in computer vision from the Hebrew University of Jerusalem in 2000. I have founded two startup companies, in areas of computer graphics and computer vision, and in 1996 I joined the founding group of 3DV Systems, developing the world's first time-of-flight (TOF) depth camera. The technology that I worked on became the basis for the depth sensors, later included in Augmented Reality headsets such as Microsoft HoloLens and Magic Leap HMDs. In 2004, I joined Microsoft Research and for 6 years I was the head of Bing Maps & Mobile research. My group developed technologies and services such as the world's first street-side service, and the popular Stroke Width Transform text detection, later included in OpenCV. In 2011 I formed a new research group at Microsoft Research, centered on augmented reality. We have developed experimental systems for an environment-aware layout of augmented reality experiences, used by several HoloLens teams and was a base for the Unity Mars product. Since 2014, I have focused on Human-Computer Interaction (HCI) and Mixed Reality (MR) and Haptics.

1 1

Fwd: [Vision-Seminar] Sunday 14/1/24 - Tomer Michaeli (Technion), room C320
by Sagie Benaim 09 Jan '24

09 Jan '24

Please note the following vision seminar that may be of interest. ---------- Forwarded message --------- From: Avital Shafran <avital.shafran(a)mail.huji.ac.il> Date: Tue, Jan 9, 2024 at 8:55 AM Subject: [Vision-Seminar] Sunday 14/1/24 - Tomer Michaeli (Technion), room C320 To: <vision-seminar(a)mail.huji.ac.il>, Tomer Michaeli < tomer.m(a)ee.technion.ac.il> Dear all, The next seminar will be given by Prof. Tomer Michaeli from the Technion. The talk will be held in room C320. Time: Sunday 14/1/24, 10:15 AM. *Title:* Quantification and Visualization of Uncertainty in Imaging Inverse Problems *Abstract:* Uncertainty quantification is crucial for the deployment of image restoration models in safety-critical domains, like biological and medical imaging. To date, methods for uncertainty visualization have mainly focused on per-pixel estimates, which provide limited information. A more natural visualization of uncertainty can be obtained from the principal components (PCs) of the posterior distribution. In this talk I will present methods for predicting the PCs of the posterior, as well as for visualizing the distribution along the space spanned by those PCs, in a single forward pass of a neural network. Our methods are both more accurate and orders of magnitude faster than the naïve approach of applying PCA to posterior samples generated by a conditional generative model. I will illustrate the effectiveness of our approaches on multiple inverse problems in imaging, including denoising, inpainting, super-resolution, colorization, and biological image-to-image translation. The talk will cover joint works with Elias Nehme, Omer Yair, and Hila Manor. *Bio:* Tomer Michaeli is an Associate Professor in the Faculty of Electrical and Computer Engineering at the Technion – Israel Institute of Technology. He completed his BSc and PhD degrees at the EE faculty of the Technion in 2005 and 2012, respectively. From 2012 to 2015 he was a postdoctoral fellow in the CS and Applied Math Department at the Weizmann Institute of Science. In 2015 he joined the Technion as a faculty member. His research lies in the fields of Computer Vision and Machine Learning. He is the recipient of several awards, among which are the Krill Prize for Excellence in Scientific Research by the Wolf foundation (2020), the Best Paper Award (Marr Prize) at ICCV 2019, and the Alon Fellowship for Outstanding Young Scientists (2017-2019). -- This group receives announcements of the talks at the Hebrew University Computer Vision Seminar. To unsubscribe from this group, send email to vision-seminar+unsubscribe(a)mail.huji.ac.il To subscribe to this group, send email to vision-seminar+subscribe(a)mail.huji.ac.il

1 0

CS Colloquium (08/01): Yuval Dagan (UC Berkeley)
by Sagie Benaim 08 Jan '24

08 Jan '24

Dear all, Next week, we have the pleasure of having Dr. Yuval Dagan give a talk in the colloquium. The seminar will be held on Monday, January 8th at 14:00. Location: C220. The title, abstract and bio appear below. Looking forward to seeing you, Sagie and Liat *Title:* Learning from dependent data and its modeling through the Ising model *Abstract:* I will present a theoretical framework for analyzing learning algorithms which rely on dependent, rather than independent, observations. While a common assumption is that the learning algorithm receives independent datapoints, such as unrelated images or texts, this assumption often does not hold. An example is data on opinions across a social network, where opinions of related people are often correlated, for example as a consequence of their interactions. I will present a line of work that models the dependence between such related datapoints using a probabilistic framework in which the observed datapoints are assumed to be sampled from some joint distribution, rather than sampled i.i.d. The joint distribution is modeled via the Ising model, which originated in the theory of Spin Glasses in statistical physics and was used in various research areas. We frame the problem of learning from dependent data as the problem of learning parameters of the Ising model, given a training set that consists of only a single sample from the joint distribution over all datapoints. We then propose using the Pseudo-MLE algorithm, and provide a corresponding analysis, improving upon the prior literature which necessitated multiple samples from this joint distribution. Our proof benefits from sparsifying a model's interaction network, conditioning on subsets of variables that make the dependencies in the resulting conditional distribution sufficiently weak. We use this sparsification technique to prove generic concentration and anti-concentration results for the Ising model, which have found applications beyond the scope of our work. Based on joint work with Constantinos Daskalakis, Anthimos Vardis Kandiros, Nishanth Dikkala, Siddhartha Jayanti, Surbhi Goel and Davin Choo. *Bio*: Yuval Dagan is a postdoctoral researcher at the Simons Institute for the Theory of Computing at UC Berkeley and at the Foundations of Data Science Institute (FODSI). He received his PhD from the Electrical Engineering and Computer Science Department at MIT, advised by Professor Constantinos Daskalakis (2018-2023). He received his Bachelor’s and Master’s degrees from the Technion, where he was advised by Professor Yuval Filmus (2011-2017). During his PhD, he received the Meta Research Fellowship in Machine Learning (2021-2022). Further, he was a visitor of the Simons Foundation at the Causality program (2022) and a research intern at Google Mountain View, hosted by Vitaly Feldman (2019). Prior to his PhD, he was a research assistant of Professor Ohad Shamir at Weizmann Institute (2018).

1 1

CS Colloquium (01/01): Yftah Ziser (Edinburgh University)
by Sagie Benaim 01 Jan '24

01 Jan '24

Dear all, Next week, we have the pleasure of having Dr. Yftah Ziser give a talk in the colloquium. The seminar will be held on Monday, January 1st at 14:00. Location: C220. The title, abstract and bio appear below. Looking forward to seeing you, Sagie and Liat *Title*: Democratising Natural Language Processing: Overcoming Language and Domain Barriers in Low-Resource Environments *Abstract:* Natural language processing (NLP) has been revolutionised in recent years to the point where it is an inseparable part of our daily lives. The transition to transformer-based models allows us to train models on vast amounts of text efficiently, proving that scale plays a crucial role in improving performance. Unfortunately, many people worldwide are marginalised from getting access to high-quality NLP models, as the language they speak and the domains they are interested in count for only a tiny fraction of current state-of-the-art models' training sets. This talk will address the challenges, approaches, and opportunities for democratising NLP across different languages and domains by developing methods to improve NLP in low-resource scenarios. I will start by discussing how we can ease distribution mismatches to improve performance using representation learning. However, as NLP models become increasingly present in our lives, improving other crucial aspects beyond their performance, such as their fairness, factuality, and our ability to understand their underlying mechanisms, is essential. Therefore, I will also discuss using spectral methods to remove information from neural networks to reduce undesired attributes, such as bias, to increase fairness where sensitive data is scarce. Finally, I will explore future directions for making these models accessible to a broader audience by improving the aspects mentioned above in low-resource scenarios. *Bio*: Yftah Ziser is a Postdoctoral Researcher at the School of Informatics at Edinburgh University, hosted by Shay Cohen. He focuses on Deep-Learning methods for dealing with the resource bottleneck, which seriously challenges the worldwide accessibility of NLP technology. His research develops methods to improve low-resource models' performance, fairness, and factuality while developing analysis methods for deepening our understanding of them. He co-organised the Domain Adaptation for NLP Workshop at EACL 2021. Before joining the University of Edinburgh, Yftah worked as a research scientist at Amazon Alexa. Yftah obtained his PhD from the Technion, where Roi Reichart advised him.

1 2

2024

2023

Colo January 2024 ----- 2024 ----- November 2024 October 2024 September 2024 August 2024 July 2024 June 2024 May 2024 April 2024 March 2024 February 2024 January 2024 ----- 2023 ----- December 2023 November 2023

Colo January 2024