The Taub Faculty of Computer Science Events and Talks

Computer systems that facilitate tight integration of data storage and processing can eliminate the "memory wall" and "power wall" bottlenecks. The memristive Memory Processing Unit (mMPU) architecture contains memory cells that can also execute logical functions, such as Memristor-Aided Logic (MAGIC) NOR gates. The mMPU supports both general-purpose computing and application-specific acceleration. This research contributes to both these aspects of the mMPU.First, we intro...
[ Full version ]

As data storage challenges grow and existing technologies approach their limits, synthetic DNA emerges as a promising storage solution due to its remarkable density and durability advantages. While cost remains a concern, emerging sequencing and synthetic technologies aim to mitigate it, yet introduce challenges such as errors in the storage and retrieval process. One crucial task in a DNA storage system is clustering numerous DNA reads into groups that represent the original i...
[ Full version ]

The memory wall bottleneck is throttling the performance of data-intensive applications as the data transfer between the processing units (e.g., CPU, GPU cores) and the memory is significantly slower than the compute itself. Therefore, emerging digital processing-in-memory systems overcome this bottleneck by performing parallel bitwise logic within the memory arrays themselves. This enables a drastic reduction in data transfer for vectored operations since the same instruction may...
[ Full version ]

Come be part of the Faculty's Capture The Flag - CTF group!! The faculty CTF group, Technipwn, holds bi-weekly meetings where we solve challenges, train for competitions and have fun The meetings are suitable for both beginners and experienced participants, and are held on Sundays at Taub 9. Organized and managed by students, and intended for all students (not only from computer science!), with and without CTF experience. The sessions include practical experience in solving chal...
[ Full version ]

Cycle-accurate simulations, frequently used by computer architects, incur substantial overheads. To mitigate this, recent virtual memory studies have adopted a lighter-weight methodology that leverages partial simulations of only the memory subsystem. This approach feeds simulation outputs into a mathematical linear model to predict execution runtimes. While this methodology accelerates the simulation process, its accuracy has traditionally been assumed rather than rigorously vali...
[ Full version ]

In this talk, we'll explore intriguing parallels between incorrect mathematical proofs and critical failures in technology, demonstrating their impact and lessons learned, like: 1. Mathematical Proofs and Social Engineering: How hand-waving a mathematical proof is similar to social engineering, and why larger hardware implementations might surprisingly be more efficient. 2. Cybersecurity and Mathematical Proofs: The surprising connection between a cyber attack that nearly caused a...
[ Full version ]

Message Passing Graph Neural Networks (MPGNNs) have emerged as the standard method for modeling complex interactions across diverse graph entities. While the theory of such models is widely investigated, their aggregation module has not received sufficient attention. Sum-based aggregators have solid theoretical foundations regarding their separation capabilities. However, practitioners often prefer using more complex aggregations and mixtures of diverse aggregations. In this resea...
[ Full version ]

First spotlight day of the semester!! Nvidia is coming to meet you in Taub Wednesday, July 10, between 12:30-2:30 PM in the Taub lobby In the program: a meeting with the recruitment team, the engineers waiting for you!!...
[ Full version ]

In this seminar, I will discuss two of my papers (published in ICLR 2023) on uncertainty estimation and class-out-of-distribution detection. We present a novel framework to benchmark the ability of image classifiers to detect class-out-of-distribution (C-OOD) instances (i.e., instances whose true labels do not appear in the training distribution) at various levels of detection difficulty. We apply this technique to ImageNet, and benchmark 500+ pretrained, publicly available, Image...
[ Full version ]

Cryptocurrencies have gained popularity in recent years. They are often implemented on one or more of the over 1000 existing blockchain networks, including the popular Bitcoin and Ethereum networks. Recently, various technologies known as blockchain interoperability have been developed to connect these different blockchains, creating an interconnected blockchain ecosystem. Interoperability refers to the ability of blockchains to share information with each other. Decentralized Exc...
[ Full version ]

In this work we introduce the problem of task assistance planning where we are given two robots Rtask and Rassist. The first robot, Rtask, is in charge of performing a given task by executing a precomputed path. The second robot, Rassist, is in charge of assisting the task performed by Rtask using on-board sensors. The ability of Rassist to provide assistance to Rtask depends on the locations of both robots. Since Rtask is moving along its path, Rassist may also need to move to pr...
[ Full version ]

Deep learning has revolutionized drug development by enhancing various stages of the process, yet traditional optimization goals often lack the sophistication required for real-world applications. This work tackles more complex and nuanced problems beyond conventional property enhancement, focusing on optimizing under patentability constraints and translating preclinical success in animals to human clinical trials. In addressing patentability, we introduce a patent loss mechanism ...
[ Full version ]

We consider the problem of approximate counting of triangles and longer fixed length cycles in undirected and directed graphs. We provide an algorithm which is faster as t, the number of copies of the searched subgraph, increases. Our running time, which significantly improves upon the state of the art (Tětek [ICALP’22]), is the same as that of multiplying an n x (n/t) matrix by an (n/t) x n matrix, up to polylogarithmic factors. Finally, we show that under popular fine-graine...
[ Full version ]

The arrival of persistent memory devices to consumer market has revived the interest in transactional durable algorithms. Persistent memory is touted as having two attributes that distinguish it from other storage technologies: fine access granularity and fast persistece.In the first part of the talk we investigate how these attributes differentiate persistent memory from block storage in the context of buffered durability – a relaxed approach that allows some progres...
[ Full version ]

Computed tomography (CT) aims to recover the volumetric three dimensional (3D) structure of heterogeneous objects. Traditionally, CT refers to a medical imaging modality. There the object, radiation source and detector array are fully controlled. This results in a linear image formation model. In contrast, we seek CT of natural objects, acquired outdoors, in an uncontrolled environment. We focus on underwater plankton and cloud droplets, which strongly affect climate. Cloud imagin...
[ Full version ]

Multivariate time series forecasting is a pivotal task in several domains, including financial planning, medical diagnostics, and climate science. This paper presents the Neural Fourier Transform (NFT) algorithm, which combines multi-dimensional Fourier transforms with Temporal Convolutional Network layers to improve both the accuracy and interpretability of forecasts. The Neural Fourier Transform is empirically validated on fourteen diverse datasets, showing superior performance ...
[ Full version ]

The annual Faculty Hackathon is underway! CS Hackathon - Doing Good Reserve the dates: June 20-21, in Taub. This year, in accordance with the order of the day, we will focus on developing technological solutions to increase mental resilience in cooperation with the Ministry of Defense's rehabilitation department, associations and initiatives. Opening of registration, development ideas, and preparation workshops - very soon. For details and more informat...
[ Full version ]

We develop a new framework, Polyform, for fast approximate matrix multiplication (AMM), through sums of sparse polynomial multiplications, using (variants of) FFT. Using this framework, we obtain new data-dependent speed-accuracy tradeoffs, which often improve on the worst-case accuracy of randomized sketching algorithms. Meanwhile, Polyform can be viewed as a cheap alternative bilinear operator to matrix multiplication in Deep Neural Networks, which is our motivating applicati...
[ Full version ]

Jumping automata are finite automata that read their input in a non-sequential manner, by allowing a reading head to “jump” between positions on the input, consuming a permutation of the input word. We argue that allowing the head to jump should incur some cost. To this end, we propose three quantitative semantics for jumping automata, whereby the jumps of the head in an accepting run define the cost of the run. The three semantics correspond to different interpretations of ju...
[ Full version ]

Neural network optimization is a challenging task, in large part because of the nonconvexity, in general, of the loss function. Various algorithms have been proposed to solve this task, but most of them focus on the convex subspaces of the loss function to avoid the challenge of finding the optimal step in concave subspaces. In addition to avoiding the concave spaces entirely (which can bear a significant computational burden!), they also simply take the minimizer of a second-d...
[ Full version ]

In this paper, we present a novel approach for test-time domain adaptation via online self-training, consisting of two components. First, we introduce a statistical framework that detects distribution shifts in the classifier's entropy values obtained on a stream of unlabeled samples. Second, we devise an online adaptation mechanism that utilizes the evidence of distribution shifts captured by the detection tool to dynamically update the classifier's parameters. The resulting adap...
[ Full version ]

Atmospheric lidars are important remote sensing tools in aerosols and climate research. A pulsed time-of-flight lidar continuously samples vertical atmospheric profiles, through day and night, yielding a spatiotemporal atmospheric map. However, lidar analysis is challenged by low signal-to-noise ratios, sunlight interference, and need for frequent calibration. We advance lidar analysis. We develop a framework for simulating realistic spatiotemporal lidar data under diverse atmosph...
[ Full version ]

DNA emerges as a promising medium for the exponential growth of digital data due to its density and durability. This study extends recent research by addressing the coverage depth problem in practical scenarios, exploring optimal error-correcting code pairings with DNA storage systems to minimize coverage depth. Conducted within random access settings, the study provides theoretical analyses and experimental simulations to examine the expectation and probability distribution of sa...
[ Full version ]

In the talk I will survey a research direction which combines coding-theory and computational complexity. This direction was introduced by Lipton, and refined by Guruswami and Smith. The goal is to construct error correcting codes which correct errors that are introduced by computationally bounded channels. This is in contrast to Hamming’s standard model which assumes a bound on the fraction of bits that a channel may flip, but allows the channel to be computationally unbounded....
[ Full version ]

In recent years, machine learning (ML) has fueled extraordinary advances in fields such as natural language processing and computer vision. These results have encouraged researchers to utilize ML in additional application domains, such as computer and networked systems, with varying degrees of success. In this talk I will focus on the prominent challenge of rate control in computer networking, and devising a practical ML toolkit for the safe deployment of deep-learning-based solut...
[ Full version ]

The CTF group of the faculty - Technipwn invites you to the semester opening lecture - a year of offensive security: attacks on OpenAI, Atlassian, Apple and more! Speaker>> Ron Massas, Vulnerability Researcher at Imperva During his work at Imperva, security researcher Ron Massas uncovered a number of fascinating vulnerabilities and hacks in a number of large companies. In the lecture, he will take us on a fascinating journey in the world of offensive security, and will present c...
[ Full version ]

In the last two decades, a main way to improve performance of computer processors has been producing processors with multiple cores, on which tasks may execute concurrently. Software is required to keep up with the hardware advances and supply concurrent algorithms for programs that run on multiple cores. This talk will focus on a fundamental building block of concurrent algorithms: concurrent data structures, designed with improved efficiency while also satisfying strong correctn...
[ Full version ]

An open day for undergraduate studies at the Technion will be held on May 31 from 10:00 am to 1:00 pm in Sarona, Tel Aviv. At the event you can meet the representatives of the faculty, consult, get an impression and get answers about the studies. Register for the open day at the ...
[ Full version ]

This week we plan to get acclimated to our seminar's new location (outside of Taub, due to construction work), in a more social atmosphere. This includes introductions of current and prospective members of the group,* and talk about plans for the semester, including theory courses scheduled. Come meet established theorists and folks interested in theory research! * if you know students doing or interested in theory who are not (or might not be) registered to this mailing list, pl...
[ Full version ]

Multi-label neural networks are important in various tasks, including safety-critical tasks. Several works show that these networks are susceptible to adversarial attacks, which can remove a target label from the predicted label list or add a target label to this list. However, no verifier can deterministically determine the list of labels for which a multi-label neural network is locally robust. The main challenge is that the complexity of the analysis increases by a factor expon...
[ Full version ]

Multi-label neural networks are important in various tasks, including safety-critical tasks. Several works show that these networks are susceptible to adversarial attacks, which can remove a target label from the predicted label list or add a target label to this list. However, no verifier can deterministically determine the list of labels for which a multi-label neural network is locally robust. The main challenge is that the complexity of the analysis increases by a factor expon...
[ Full version ]

A meeting of those interested in graduate studies at the Taub Faculty of Computer Science at the Technion will be held on Wednesday, May 22, 2024, 1:00 p.m., room 337 Outstanding bachelor's graduates from all universities! This is your opportunity to participate and be impressed by the Faculty of Computer Science at the Technion, Meet faculty members and graduate students and hear fascinating lectures. Event schedule: 13:00-13:30 Opening remarks: The words of t...
[ Full version ]

Payment Channel Networks (PCNs) are a leading method to scale the transaction throughput in cryptocurrencies. Two participants can use a bidirectional payment channel for making multiple mutual payments without committing them to the blockchain. Opening a payment channel is a slow operation that involves an on-chain transaction locking a certain amount of funds. These aspects limit the number of channels that can be opened or maintained. Users may route payments through a multi-ho...
[ Full version ]

In the past few years, software has been at the heart of many applications ranging from appliances to virtual services. Helping software developers to write better code is a crucial task. In parallel, recent developments in machine learning and deep learning in particular have shown great promise in many fields, and code-related tasks in particular. The main challenge is how to represent code in a way that can be used by deep learning models effectively. While code can be treated ...
[ Full version ]

We introduce and study jumping automata over infinite words, a fascinating twist on traditional finite automata. These machines read their input in a non-consecutive manner, defying conventional word order. We explore three distinct semantics: one ensuring every letter is accounted for, another permitting word permutation within fixed windows, and a third allowing permutation within windows of an existentially-quantified bound. Our work covers expressiveness, closure properties, a...
[ Full version ]

Methods in question-answering (QA) that transform texts detailing processes into an intermediate code representation, subsequently executed to generate a response to the presented question, have demonstrated promising results in analyzing scientific texts that describe intricate processes. The limitations of these existing text-to-code models are evident when attempting to solve QA problems that require knowledge beyond what is presented in the input text. We propose a novel domai...
[ Full version ]

Distributed lending protocols are major financial mechanisms that match borrowers and lenders without the need for intermediaries. Smart contracts deployed on blockchains ensure the security of such loans. Borrowers mitigate the risk of default by providing collateral in the form of cryptocurrencies or by limiting the loan duration. In the first part of the talk, we describe various loan types, major platforms, and financial models that allow them. In the second part of the talk,...
[ Full version ]

In biological imaging, fast acquisition of depth information is crucial e.g. for accurate 3D tracking of sub-cellular elements and for 3D super-resolution. In the first part of this talk, we present a series of works enhancing the success of snapshot depth sensing in the revolutionary field of single-molecule localization microscopy (Nobel Prize in Chemistry 2014). Specifically, we present an approach for jointly designing the “optics” of the microscope and the 3D reconstructi...
[ Full version ]

The standard formulation of Markov decision processes (MDPs) assumes that the agent's decisions are executed immediately. However, in numerous realistic applications such as robotics or healthcare, actions are performed with a delay whose value can even be stochastic. In this work, we introduce stochastic delayed execution MDPs, a new formalism addressing random delays without resorting to state augmentation. We show that given observed delay values, it is sufficient to perform a ...
[ Full version ]

Graduates of the faculty are invited with their children to a day of lectures, workshops and information regarding registration and admission - for children; and a nostalgic tour (for parents) Monday, April 15, starting at 10:00 at the Technion In the program: 1. The wonders of infinity - the Faculty of Mathematics 2. What is the Internet of Things? - Faculty of Computer Science 3. Games and Virtual Reality - Fac...
[ Full version ]

Due to their increasing aggressiveness, recent congestion control algorithms (CCAs) can quickly starve standard TCP flows in their shared router queues. Existing solutions based on fair queueing are not scalable enough, and those based on admission control do not fit all CCAs. Independently, building on the popularization of machine learning, recent papers have designed several CCA classifiers.In this seminar, we introduce a buffer isolation mode where incoming flows first...
[ Full version ]

Graduate Seminar Data augmentation has become an integral part of deep learning, as it is known to improve the generalization capabilities of neural networks.Since the most effective set of image transformations differs between tasks and domains, automatic data augmentation search aims to alleviate the extreme burden of manually finding the optimal image transformations. However, current methods are not able to jointly optimize all degrees of freedom: (1)...
[ Full version ]

In the landscape of data stream processing, the challenges posed by online problems are of paramount importance. This paper introduces a pioneering method tailored to address these challenges within the context of data streams. While our initial focus centered on Complex Event Processing (CEP) problems, it is essential to underscore the versatility of our approach, as it is equally applicable to a diverse range of online problems. To the best of our knowledge, there exists no comp...
[ Full version ]

Assume that you have a huge graph, where edges and vertices may fail, and you want to answer questions about this graph quickly, without inspecting the whole graph. A fault-tolerant labeling scheme allows us to do just that. It is a distributed data structure, in which each vertex and edge is assigned a short label, such that given the labels of a pair of vertices s,t, and a set of failures F, you can answer questions about s and t in G\F. For example, determine if s and t are con...
[ Full version ]

Language Models (LMs) excel in many tasks, but understanding discourse – how sentences connect to form coherent text – remains a challenge. This is especially true for smaller models aiming to match the abilities of their larger counterparts in handling long and complex inputs. To address this, we introduce DEPTH, a new encoder-decoder model designed to foster robust discourse-level representations during the pre-training phase. DEPTH uniquely combines hierarchical sentence re...
[ Full version ]

Between training and inference, lies a growing class of AI problems that involve fast optimization of a pre-trained model for a specific inference task. These are not pure “feed-forward” inference problems applied to a pre-trained model, because they involve some non-trivial inference-time optimization beyond what the model was trained for; neither are they training problems, because they focus on a specific input. These compute-heavy inference workflows raise new challenges i...
[ Full version ]

The 12th CS Research Day for graduate studies will be held on Wednesday, April 03, 2024 between 12:30-14:30, at the lobby of the CS Taub Building. Research Day events are opportunity for our graduate students to expose their researches using posters and presentations to CS faculty and all degrees students, Technion distinguished representatives and to high-ranking delegates from the hi-tech leading industry companies in Israel and abroad. The participating researches will be on...
[ Full version ]

One of the major open problems in complexity theory is proving super-logarithmic lower bounds on the depth of circuits. Karchmer, Raz, and Wigderson (Computational Complexity 5(3/4), 1995) suggested approaching this problem by proving that depth complexity of a composition of two functions is roughly the sum of their individual depth complexities. They showed that the validity of this conjecture would imply the desired lower bounds.The intuition that underlies the KRW conj...
[ Full version ]

In the past year, numerous companies have incorporated Generative AI (GenAI) capabilities into new and existing applications, forming interconnected Generative AI (GenAI) ecosystems consisting of applications powered by GenAI services.While ongoing research highlighted risks associated with the GenAI layer of agents (e.g., dialog poisoning, membership inference, prompt leaking, jailbreaking), a critical question emerges: Can attackers develop malware to exploit the GenAI c...
[ Full version ]

Foundation models that connect vision and language have recently shown great promise for a wide array of tasks such as text-to-image generation. Significant attention has been devoted towards utilizing the visual representations learned from these powerful vision and language models. In this talk, I will present an ongoing line of research that focuses on the other direction, aiming at understanding what knowledge language models acquire through exposure to images during pretraini...
[ Full version ]

In various applications, such as virtual reality and gaming, simulating the deformation of soft tissues in the human body during interactions with external objects is essential. Traditionally, Finite Element Methods (FEM) have been employed for this purpose, but they tend to be slow and resource-intensive. In this paper, we propose a unified representation of human body shape and soft tissue with a data-driven simulator of non-rigid deformations. This approach enables rapid simula...
[ Full version ]

Engineering the Future of Blockchain 1.4.23 | 12:30-14:30 | The graduate students' lounge, Taub, 2nd floor 12:30 - Mingling and Snacks 13:00 - Oded Naor, Ph.D., Product Manager - Btockchain reinvented: a deep dive into StarkWare's game-changing sotutions 13:30 - Noa Oved, MSC, Software Team Lead - Coding the future: Transforming algorithmic insights into real-wortd sotutions 14:00- Q&A Plese RSVP at...
[ Full version ]

An open day for studies at the Technion will be held on Thursday, March 28, starting at 09:30 For details and registration, go to the registration and admission website at the link ...
[ Full version ]

In machine learning, there has been considerable interest over the past decade in understanding the ability of optimization algorithms to generalize—namely, to produce solutions (models) that extend well to unseen data—particularly in the context of overparameterized problems. I will survey several recent theoretical studies that explore generalization in classical convex optimization, that reveal intriguing behavior of common optimization methods and shed some light on the c...
[ Full version ]

I will explain how successful innovations in software can usually be traced to just one or two "concepts" that offer new scenarios that, with seemingly small shifts, radically change how an application is used. I give examples from apps such as Zoom, WhatsApp and Calendly. I explain how concepts, each with their own characteristic scenarios, can be composed to form apps. I explain how this idea can be used to make software more usable, modular and consistent. More information at:...
[ Full version ]

Age of Information (AoI) is a recently proposed performance metric that captures the freshness of the information from the perspective of the application. AoI measures the time that elapsed from the moment that the most recently received packet was generated to the present time. In this talk, we explore the AoI optimization problem in wireless networks.We start by considering a wireless network with a number of nodes transmitting information to a base station and develop l...
[ Full version ]

Intel's tech experience is coming to campus! Ready for the most innovative technological picnic you've ever seen? We have loaded technological tools and our greatest minds to the track and we are on our way to you Wednesday 27.3 | Mayer building, 3rd floor 10:00 - AR | VR | HR Come and get to know us and our technologies 12:30 - FPGA MicroPython The number of places is limited, register at the ...
[ Full version ]

We propose a new approach for generative modeling based on training a neural network to be idempotent. An idempotent operator is one that can be applied sequentially without changing the result beyond the initial application, namely f(f(z))=f(z). The proposed model f is trained to map a source distribution (e.g, Gaussian noise) to a target distribution (e.g. realistic images) using the following objectives: (1) Instances from the target distribution should map to themselves, namel...
[ Full version ]

In this lecture we will explore collaborative caching algorithms in order to boost the effectiveness of caches in a distributed storage system. I'll introduce a scheme that partitions each node’s cache into two conceptual regions: an egoistic area whose goal is to contain the most valuable data for the node that owns the cache, and an altruistic area whose goal is to contain the most valuable data for the system as a whole. Each node’s division between these two regions is dyn...
[ Full version ]

We propose a new way to explain and to visualize neural network classification through a decomposition-based explainable AI. Instead of providing an explanation heatmap, our method yields a decomposition of the image into class-agnostic and class-distinct parts, with respect to the data and chosen classifier. Following a fundamental signal processing paradigm of analysis and synthesis, the original image is the sum of the decomposed parts. We thus obtain a radically different way ...
[ Full version ]

A finite metric space $(X,d)$ on a set of points $X$ is just the shortest path metric $d$ on a positively weighted graph $G=(X,E)$. In the online setting, the vertices of the input finite metric space $(X,d)$ are exposed one by one, together with their distances $d(*,*)$to the previously exposed vertices. The goal is to embed (map) $X$ into a given host metric space $(H,d_H)$ (finite or not) and so to distort the distances as little as possible (distortion is the worst cas...
[ Full version ]

You are invited to the spotlight day of the Mobileye company at the Technion Wednesday 20.3 | 12:30 | Taub Building...
[ Full version ]

The surveillance economy, where tracking and collecting data on uses for the purpose of advertising and other actions, is central to much of the money-making enterprises of the modern technology ecosystem. Due to regulations and other forces, some of the largest companies, such as Google and Apple, have prioritized mechanisms for users to better manage and receive information about the kinds of data that is being collected about them. In this talk, I will explore how effective the...
[ Full version ]

Yahoo holds a dedicated meeting with graduate students Tuesday, March 19 at 12:30 p.m. at the CS Grads Club In the program: an introduction to Yahoo's research in Israel and the summer internship program. Register to the event here waiting for you!!...
[ Full version ]

Join us for a lecture and pizza! Cyber Security Challenges In The Modern Era by Yarden Zuckerman, SW Security Manager, Nvidia Monday | March 18, 2024 | 18:00 p.m. | Piano Auditorium Taub Building ...
[ Full version ]

I'll present our paper about the role that null messages play in synchronous systems with and without failures. Our work provides necessary and sufficient conditions on the structure of protocols for information transfer and coordination there. We start by introducing a new and more refined definition of null messages. A generalization of message chains that allow these null messages is provided and is shown to be necessary and sufficient for information transfer in reliable syste...
[ Full version ]

Differential Privacy (DP) is a mathematical framewirk for ensuring the privacy of individuals in a dataset. Roughly speaking, it guarantees that privacy is protected in data analysis by ensuring that the output of an analysis does not reveal sensitive information about any specific individual, regardless of whether their data is included in the dataset or not.This talk presents a unified framework for characterizing both pure and approximate differentially private learnabi...
[ Full version ]

To deploy and operate deep neural models in production, the quality of their predictions, which might be contaminated benignly or manipulated maliciously by input distributional deviations, must be monitored and assessed. Specifically, we study the case of monitoring the healthy operation of a deep neural network (DNN) receiving a stream of data, with the aim of detecting input distributional deviations over which the quality of the network’s predictions is potentially damaged. ...
[ Full version ]

Multi-rotor aerial autonomous vehicles (MAVs) primarily rely on vision for navigation purposes. However, visual localization and odometry techniques suffer from poor performance in low or direct sunlight, a limited field of view, and vulnerability to occlusions. Acoustic sensing can serve as a complementary or even alternative modality for vision in many situations, and it also has the added benefits of lower system cost and energy footprint, which is especially important for micr...
[ Full version ]

We propose that spaceborne polarimetric imagers can be calibrated, or self-calibrated using zodiacal light (ZL). ZL is created by a cloud of interplanetary dust particles. It has a significant degree of polarization in a wide field of view. From space, ZL is unaffected by terrestrial disturbances. ZL is insensitive to the camera location, so it is suited for simultaneous cross-calibration of satellite constellations. ZL changes on a scale of months, thus being a quasi-constant tar...
[ Full version ]

I will share some of my learnings from working on problems on the intersection of Distributed Computing and Cryptography.On the one hand, I will show how some cryptographic protocols (MPC and DKG) can be improved by using distributed computing counterparts for notions such as zero knowledge proofs and proofs of knowledge. On the other hand, I will show how distributed computing protocols (ABA) can be improved by carefully using notions of binding from cryptography. One rec...
[ Full version ]

Most works on modeling the conversation history in Conversational Question Answering (CQA) report a single main result on a common CQA benchmark. While existing models show impressive results on CQA leaderboards, it remains unclear whether they are robust to shifts in setting (sometimes to more realistic ones), training data size (e.g. from large to small sets) and domain. In this work, we design and conduct the first large-scale robustness study of history modeling approaches for...
[ Full version ]

Nvidia invites you to a virtual spotlight day where the company's engineers will talk about the different groups and open positions Sunday, March 10, from 18:00 to 19:30 To register for the event here...
[ Full version ]

Cadence company is coming to a spotlight day at the faculty Wednesday 6/3 | 12:30-14:30 | Lobby Taub In the program: a meeting with the recruitment teams, the engineers for a 1:1 conversation about employment opportunities and tips for writing a report. And of course merch and sweets. waiting for you!!...
[ Full version ]

We design a new stochastic first-order algorithm for approximately solving matrix games as well as the more general problem of minimizing the maximum of smooth convex functions. Our central tool is ball oracle acceleration: a technique for minimizing any convex function with a small number of calls to a ball oracle that minimizes the same function restricted to a small ball around the query point. To design an efficient ball oracle for our problems of interest we leverage stochast...
[ Full version ]

Early time classification algorithms aim to label a stream of features without processing the full input stream, while maintaining accuracy comparable to that achieved by applying the classifier to the entire input. In this paper, we introduce a statistical framework that can be applied to any sequential classifier, formulating a calibrated stopping rule. This data-driven rule attains finite-sample, distribution-free control of the accuracy gap between full and early-time classifi...
[ Full version ]

You have been accepted as a student! What now? The student community at the SHE S faculty invites you to a meeting on: First time student job: about the opportunities, challenges and skills you acquire in your first job Monday, 4/3 at 6:30 pm in the piano auditorium Please register in advance: here Speaker: Moore Philo, graduate of the faculty and developer at Amazon The meeting is suitable for thos...
[ Full version ]

We study the problem of computing an embedding of the tuples of a relational database in a manner that is extensible to dynamic changes of the database. In this problem, the embedding should be stable in the sense that it should not change on the existing tuples due to the embedding of newly inserted tuples (as database applications might already rely on existing embeddings); at the same time, the embedding of all tuples, old and new, should retain high quality. This task is chall...
[ Full version ]

Spotlight day for Nvidia at the Technion on February 28, 2024 Nvidia is coming to meet the students of the Faculty of Computer Science at the Technion! Come and meet the company's engineers face to face Wednesday February 28, 2024 | 12:30-14:30 | Taub lobby, floor 0...
[ Full version ]

Theory Seminar: An f-edge fault-tolerant distance sensitivity oracle (f-DSO) is a data-structure that, when queried with two vertices (s, t) and a set F of at most f edges of a graph G with n vertices, returns an estimate tilde{d}(s,t,F) of the distance d(s,t,F) from s to t in G – F. The oracle has stretch alpha if the estimate satisfies d(s,t,F) le tilde{d}(s,t,F) le alpha cdot d(s,t,F) . In the last two decades, extensive research has focused on developing efficient f-DSOs. Th...
[ Full version ]

Locks are frequently used in concurrent systems to simplify code and ensure safe access to contended parts of memory. However, they are also known to cause bottlenecks in concurrent code, leading practitioners and theoreticians to sometimes opt for more intricate lock-free implementations. In this talk, I’ll show that, despite the seeming contradiction, it is possible to design practically and theoretically efficient lock-free locks; I'll present a lock-free lock algorithm with ...
[ Full version ]

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. Later, we expl...
[ Full version ]

In the last decade, DNA-based storage systems emerged as a potential data archival solution due to their high data density and durability. This research delves into intrinsic error characteristics of DNA storage systems to devise robust coding strategies and innovative algorithms for enhanced reliability, efficiency, scalability, and cost-effectiveness. The research propels DNA storage feasibility while contributing to foundational theory.The work analyzes combinatorial st...
[ Full version ]

High-dimensional data is increasingly available in diverse applications, ranging from images to shapes represented as point clouds. Such data raises novel questions and offers a unique opportunity to study them by developing new machine-learning tools. While the analysis of an individual sample may be challenging, leveraging the power of the data collection can be effective in tackling complex tasks. This talk delves into the challenges associated with studying such data, particul...
[ Full version ]

We show that every language in NP has an Interactive Oracle Proof (IOP) with inverse polynomial soundness error and small query complexity. This achieves parameters that surpass all previously known PCPs and IOPs. Specifically, we construct an IOP with perfect completeness, soundness error 1/n, round complexity O(loglog n), proof length poly(n) over an alphabet of size O(n), and query complexity O(loglog n). This is a step forward in the quest to establish the sliding-scale conjec...
[ Full version ]

This talk focuses on the distributed task of constructing an approximate \emph{maximum weight independent set (MWIS)}. Specifically, we are interested in deterministic CONGEST algorithms whose approximation guarantees are expressed as a function of the graph's \emph{arboricity} $\alpha$.Generally speaking, efficient deterministic non-trivial approximation algorithms for MWIS were not known until the recent breakthrough of Faour et al.~[SODA 2023] that obtained an $O(\Delta...
[ Full version ]

In clinical settings, a significant portion of ECG data is typically available in printed form, and the most convenient means of digitizing this information involves utilizing a mobile device. Despite notable progress in AI-based techniques for paper-based 12-lead ECG analysis, their adoption in clinical practice remains limited primarily due to challenges such as inadequate accuracy in clinical settings and a restricted ability to diagnose various cardiac conditions. Our objectiv...
[ Full version ]

Suppose that n forecasting experts (or functions) are providing daily rain/no-rain predictions, and the best among them is mistaken in at most k many days. For how many days will a person allowed to observe the predictions, but has no weather-forecasting prior knowledge, mispredict?In this talk, we will discuss how such classical problems can be reduced to calculating the (average) depth of binary trees, by using newly introduced complexity measures (aka dimensions) of the...
[ Full version ]

With the rapid increase of storage demands and working sets of modern mobile apps, maintaining high I/O performance in mobile SSDs under strict resource constraints is challenging. The Flash Translation Layer (FTL) must increase the capacity of the Logical-To-Physical (L2P) address translation cache to keep up with the new workloads, but it comes at the cost of scaling the on-die SRAM, resulting in higher chip area, power consumption, and costs.In this talk, I will present...
[ Full version ]

Come be part of a new Capture The Flag-CTF group at the Faculty The meetings are held every Tuesday at 18:30 at Taub 9 and include guest lectures and practical experience in solving challenges. Everybody is invited! Beginners and experienced For details: Technionctf.com This week - February 13, 2024: 18:30 | Taub 2 | Omar Atias - security researcher, lecturer at BlackHat USA & DEFCON The price of con...
[ Full version ]

QUIC is an emerging transport protocol, offering multiple advantages over TCP. Yet, to fully unleash QUIC’s potential, a paradigm shift is needed in existing network infrastructure. We propose a novel 0-RTT-aware load balancing algorithm. 0-RTT is crucial for web performance, particularly on mobile networks. Our load balancing algorithm ensures 0-RTT while maintaining near-optimal load balancing performance.Through extensive simulations, using both synthetic and real-wor...
[ Full version ]

You are invited to Final spotlight day Wednesday 07.02.2024 | 12:30-14:30 | Visitor Center Auditorium 012, Floor 0 12:30 - Come meet engineers, researchers and the recruitment team at Final, and get to know the day-to-day life at Final. 13:15 - Meeting on options, probabilities and the world of algorithm trading | Noam Horowitz - researcher at Final To register for the lecture click ...
[ Full version ]

Come be part of a new Capture The Flag-CTF group at the Faculty The meetings are held every Tuesday at 18:30 at Taub 9 and include guest lectures and practical experience in solving challenges. Everybody is invited! Beginners and experienced For details: Technionctf.com The week of February 6, 2024: Beginners: Forensics & Networks | Taub 9 experienced: Challenges from 2023 LA CTF | Tau...
[ Full version ]

Come be part of a new Capture The Flag-CTF group at the Faculty The meetings are held every Tuesday at 18:30 at Taub 9 and include guest lectures and practical experience in solving challenges. Everybody is invited! Beginners and experienced For details: Technionctf.com The week of February 6, 2024: Beginners: Forensics & Networks | Taub 9 experienced: Challenges from 2023 LA CTF | Taub 8...
[ Full version ]

Many object-oriented applications in algorithm design rely on objects never changing during their lifetime. This is often tackled by marking object references as read-only, e.g., using the const keyword in C++. In other languages like Python or Java where such a concept does not exist, programmers rely on best practices that are entirely unenforced. While reliance on best practices is obviously too permissive, const-checking is too restrictive: it is possible for a method to mutat...
[ Full version ]

The visual object category reports of artificial neural networks (ANNs) are notoriously sensitive to tiny, adversarial image perturbations. Because human category reports (aka human percepts) are thought to be insensitive to those same small-norm perturbations — and locally stable in general — this argues that ANNs are incomplete scientific models of human visual perception. Consistent with this, we show that when small-norm image perturbations are generated by standard ANN mo...
[ Full version ]

We study secure multiparty computation in the asynchronous setting with perfect security and optimal resilience (less than one-fourth of the participants are malicious). It has been shown that every function can be computed in this model [Ben-OR, Canetti, and Goldreich, STOC'1993]. Despite 30 years of research, all protocols in the asynchronous setting require $\Omega(n^2C)$ communication complexity for computing a circuit with $C$ multiplication gates. In contrast, for nearly 15...
[ Full version ]

We study several graph layout problems with a min max objective. Here, given a graph we wish to find a linear ordering of the vertices that minimizes some worst case objective over the natural cuts in the ordering; which separate an initial segment of the vertices from the rest. A prototypical problem here is cutwidth, where we want to minimize the maximum number of edges crossing a cut. The only known algorithm here is by [Leighton-Rao J.ACM 99] based on recursively partitioning ...
[ Full version ]

We study several graph layout problems with a min max objective. Here, given a graph we wish to find a linear ordering of the vertices that minimizes some worst case objective over the natural cuts in the ordering; which separate an initial segment of the vertices from the rest. A prototypical problem here is cutwidth, where we want to minimize the maximum number of edges crossing a cut. The only known algorithm here is by [Leighton-Rao J.ACM 99] based on recursively partitioning ...
[ Full version ]

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. Atta...
[ Full version ]

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 con...
[ Full version ]

Text-to-image diffusion models (T2I) use a latent representation of a text prompt to guide the image generation process. However, the encoder that produces the text representation is largely unexplored. We propose the Diffusion Lens, a method for analyzing the text encoder of T2I models by generating images from its intermediate representations. Using the Diffusion Lens, we perform an extensive analysis of two recent T2I models.We find that the text encoder gradually build...
[ Full version ]

Hyperproperties are system properties that relate multiple execution traces to one another. Hyperproperties are essential to express a wide range of system requirements such as information flow and security policies; epistemic properties like knowledge in multi-agent systems; fairness; and robustness. With the aim of verifying program correctness, the two major challenges are (1) providing a specification language that can precisely express the desired properties; and (2) providin...
[ Full version ]

Stochastic convex optimization is one of the most well-studied models for learning in modern machine learning. Nevertheless, a central fundamental question in this setup remained unresolved: How many data points must be observed so that any empirical risk minimizer (ERM) shows good performance on the true population? This question was proposed by Feldman who proved that Ω(\frac{d}{ϵ} + \frac{1}{ϵ^2}) data points are necessary (where d is the dimension and ε > 0 is the accurac...
[ Full version ]

The Technion CTF Team opens at the Faculty We invite you to join the Capture The Flag - CTF meetings. CTF is a cyber challenge competition and information security on the topics: cryptography, reverse engineering, forensics, web, etc. The meetings will include guest lectures and practical experience in solving challenges. Beginner and experienced students are welcome to join. The first introductory meeting will be held on Tuesday, January 23 at 6:30 pm at Taub 9. In the f...
[ Full version ]

In this talk we consider the problem of manifold reconstruction from oriented point clouds for embedded manifolds of co-dimension larger than one. Using the framework of Poisson Surface Reconstruction, and formulating the problem in the language of alternating products, we show that the earlier approach for reconstructing hyper-surfaces extends to general manifolds, at the cost of replacing a quadratic energy with a multi-quadratic energy. We provide an efficient iterative hierarc...
[ Full version ]

Vizing’s Theorem provides an algorithm that edge colors any graph of maximum degree Δ using Δ+1 colors, which is necessary for some graphs, and at most one higher than necessary for any graph. In online settings, the trivial greedy algorithm requires 2Δ-1 colors, and Bar-Noy, Motwani and Naor in the early 90s showed that this is best possible, at least in the low-degree regime. In contrast, they conjectured that for graphs of superlogarithmic-in-n maximum degree, much better ...
[ Full version ]

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 ...
[ Full version ]

12-lead electrocardiogram (ECG) recordings can be collected in any clinic and the interpretation is performed by a clinician. Modern machine learning tools may make them automatable. However, a large fraction of 12-lead ECG data is still available in printed paper or image only and comes in various formats. To digitize the data, smartphone cameras can be used. Nevertheless, this approach may introduce various artifacts and occlusions into the obtained images.Here, I will p...
[ Full version ]

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 o...
[ Full version ]