Showing 1–10 of 10 results for author: Meir, S

First-Passage Approach to Optimizing Perturbations for Improved Training of Machine Learning Models

Authors: Sagi Meir, Tommer D. Keidar, Shlomi Reuveni, Barak Hirshberg

Abstract: Machine learning models have become indispensable tools in applications across the physical sciences. Their training is often time-consuming, vastly exceeding the inference timescales. Several protocols have been developed to perturb the learning process and improve the training, such as shrink and perturb, warm restarts, and stochastic resetting. For classifiers, these perturbations have been sho… ▽ More Machine learning models have become indispensable tools in applications across the physical sciences. Their training is often time-consuming, vastly exceeding the inference timescales. Several protocols have been developed to perturb the learning process and improve the training, such as shrink and perturb, warm restarts, and stochastic resetting. For classifiers, these perturbations have been shown to result in enhanced speedups or improved generalization. However, the design of such perturbations is usually done ad hoc by intuition and trial and error. To rationally optimize training protocols, we frame them as first-passage processes and consider their response to perturbations. We show that if the unperturbed learning process reaches a quasi-steady state, the response at a single perturbation frequency can predict the behavior at a wide range of frequencies. We employ this approach to a CIFAR-10 classifier using the ResNet-18 model and identify a useful perturbation and frequency among several possibilities. Our work allows optimization of perturbations for improving the training of machine learning models using a first-passage approach. △ Less

Submitted 13 March, 2025; v1 submitted 6 February, 2025; originally announced February 2025.

Distributed Download from an External Data Source in Faulty Majority Settings

Authors: John Augustine, Soumyottam Chatterjee, Valerie King, Manish Kumar, Shachar Meir, David Peleg

Abstract: We extend the study of retrieval problems in distributed networks, focusing on improving the efficiency and resilience of protocols in the \emph{Data Retrieval (DR) Model}. The DR Model consists of a complete network (i.e., a clique) with $k$ peers, up to $βk$ of which may be Byzantine (for $β\in [0, 1)$), and a trusted \emph{External Data Source} comprising an array $X$ of $n$ bits ($n \gg k$) th… ▽ More We extend the study of retrieval problems in distributed networks, focusing on improving the efficiency and resilience of protocols in the \emph{Data Retrieval (DR) Model}. The DR Model consists of a complete network (i.e., a clique) with $k$ peers, up to $βk$ of which may be Byzantine (for $β\in [0, 1)$), and a trusted \emph{External Data Source} comprising an array $X$ of $n$ bits ($n \gg k$) that the peers can query. Additionally, the peers can also send messages to each other. In this work, we focus on the Download problem that requires all peers to learn $X$. Our primary goal is to minimize the maximum number of queries made by any honest peer and additionally optimize time. We begin with a randomized algorithm for the Download problem that achieves optimal query complexity up to a logarithmic factor. For the stronger dynamic adversary that can change the set of Byzantine peers from one round to the next, we achieve the optimal time complexity in peer-to-peer communication but with larger messages. In broadcast communication where all peers (including Byzantine peers) are required to send the same message to all peers, with larger messages, we achieve almost optimal time and query complexities for a dynamic adversary. Finally, in a more relaxed crash fault model, where peers stop responding after crashing, we address the Download problem in both synchronous and asynchronous settings. Using a deterministic protocol, we obtain nearly optimal results for both query complexity and message sizes in these scenarios. △ Less

Submitted 27 December, 2024; originally announced December 2024.

Comments: 39 pages

Byzantine Resilient Computing with the Cloud

Authors: John Augustine, Jeffin Biju, Shachar Meir, David Peleg, Srikkanth Ramachandran, Aishwarya Thiruvengadam

Abstract: We study a framework for modeling distributed network systems assisted by a reliable and powerful cloud service. Our framework aims at capturing hybrid systems based on a point to point message passing network of machines, with the additional capability of being able to access the services of a trusted high-performance external entity (the cloud). We focus on one concrete aspect that was not studi… ▽ More We study a framework for modeling distributed network systems assisted by a reliable and powerful cloud service. Our framework aims at capturing hybrid systems based on a point to point message passing network of machines, with the additional capability of being able to access the services of a trusted high-performance external entity (the cloud). We focus on one concrete aspect that was not studied before, namely, ways of utilizing the cloud assistance in order to attain increased resilience against Byzantine behavior of machines in the network. Our network is modeled as a congested clique comprising $k$ machines that are completely connected to form a clique and can communicate with each other by passing small messages. In every execution, up to $βk$ machines (for suitable values of $β\in [0, 1)$) are allowed to be Byzantine, i.e., behave maliciously including colluding with each other, with the remaining $γk$ or more machines being \emph{honest} (for $γ=1-β$). Additionally, the machines in our congested clique can access data through a trusted cloud via queries. This externality of the data captures many real-world distributed computing scenarios and provides a natural context for exploring Byzantine resilience for essentially all conceivable problems. Moreover, we are no longer bound by the usual limits of $β< 1/3$ or even $β< 1/2$ that are typically seen in Byzantine Agreement. We focus on a few fundamental problems. We start with the ${\textsf{Download}}$ problem, wherein the cloud stores $n$ bits and these $n$ bits must be downloaded to all of the $k$ machines. In addition to ${\textsf{Download}}$, we also consider the problem of computing the ${\textsf{Disjunction}}$ and ${\textsf{Parity}}$ of the bits in the cloud. We study these problems under several settings comprising various $β$ values and adversarial capabilities. △ Less

Submitted 28 September, 2023; originally announced September 2023.

Comments: 54 pages

Modulating Hierarchical Self-Assembly In Thermoresponsive Intrinsically Disordered Proteins Through High-Temperature Incubation Time

Authors: Vaishali Sethi, Dana Cohen-Gerassi, Sagi Meir, Max Ney, Yulia Shmidov, Gil Koren, Lihi Adler-Abramovich, Ashutosh Chilkoti, Roy Beck

Abstract: The cornerstone of structural biology is the unique relationship between protein sequence and the 3D structure at equilibrium. Although intrinsically disordered proteins (IDPs) do not fold into a specific 3D structure, breaking this paradigm, some IDPs exhibit large-scale organization, such as liquid-liquid phase separation. In such cases, the structural plasticity has the potential to form numero… ▽ More The cornerstone of structural biology is the unique relationship between protein sequence and the 3D structure at equilibrium. Although intrinsically disordered proteins (IDPs) do not fold into a specific 3D structure, breaking this paradigm, some IDPs exhibit large-scale organization, such as liquid-liquid phase separation. In such cases, the structural plasticity has the potential to form numerous self-assembled structures out of thermal equilibrium. Here, we report that high-temperature incubation time is a defining parameter for micro and nanoscale self-assembly of resilin-like IDPs. Interestingly, high-resolution scanning electron microscopy micrographs reveal that an extended incubation time leads to the formation of micron-size rods and ellipsoids that depend on the amino acid sequence. More surprisingly, a prolonged incubation time also induces amino acid composition-dependent formation of short-range nanoscale order, such as periodic lamellar nanostructures. We, therefore, suggest that regulating the period of high-temperature incubation, in the one-phase regime, can serve as a unique method of controlling the hierarchical self-assembly mechanism of structurally disordered proteins. △ Less

Submitted 30 November, 2023; v1 submitted 1 September, 2023; originally announced September 2023.

Comments: 27pages, 8 figures

Intramolecular Structural Heterogeneity altered by Long-range Contacts in an Intrinsically Disordered Protein

Authors: Gil Koren, Sagi Meir, Lennard Holschuh, Haydyn D. T. Mertens, Tamara Ehm, Nadav Yahalom, Adina Golombek, Tal Schwartz, Dmitri I. Svergun, Omar A. Saleh, Joachim Dzubiella, Roy Beck

Abstract: Short-range interactions and long-range contacts drive the 3D folding of structured proteins. The proteins' structure has a direct impact on their biological function. However, nearly 40% of the eukaryotes proteome is composed of intrinsically disordered proteins (IDPs) and protein regions that fluctuate between ensembles of numerous conformations. Therefore, to understand their biological functio… ▽ More Short-range interactions and long-range contacts drive the 3D folding of structured proteins. The proteins' structure has a direct impact on their biological function. However, nearly 40% of the eukaryotes proteome is composed of intrinsically disordered proteins (IDPs) and protein regions that fluctuate between ensembles of numerous conformations. Therefore, to understand their biological function, it is critical to depict how the structural ensemble statistics correlate to the IDPs' amino acid sequence. Here, using small-angle x-ray scattering (SAXS) and time-resolved Förster resonance energy transfer (trFRET), we study the intra-molecular structural heterogeneity of the neurofilament low intrinsically disordered tail domain (NFLt). Using theoretical results of polymer physics, we find that the Flory scaling exponent of NFLt sub-segments correlates linearly with their net charge, ranging from statistics of ideal to self-avoiding chains. Surprisingly, measuring the same segments in the context of the whole NFLt protein, we find that regardless of the peptide sequence, the segments' structural statistics are more expanded than when measured independently. Our findings show that while polymer physics can, to some level, relate the IDP's sequence to its ensemble conformations, long-range contacts between distant amino acids play a crucial role in determining intra-molecular structures. This emphasizes the necessity of advanced polymer theories to fully describe IDPs ensembles with the hope it will allow us to model their biological function. △ Less

Submitted 9 June, 2023; v1 submitted 4 December, 2022; originally announced December 2022.

Polarization dynamics of ultrafast solitons

Authors: Avi Klein, Sara Meir, Hamootal Duadi, Arjunan Govindarajan, Moti Fridman

Abstract: We study the polarization dynamics of ultrafast solitons in mode-locked fiber lasers. We find that when a stable soliton is generated, it's state-of-polarization shifts toward a stable state and when the soliton is generated with excess power levels it experiences relaxation oscillations in its intensity and timing. On the other hand, when a soliton is generated in an unstable state-of-polarizatio… ▽ More We study the polarization dynamics of ultrafast solitons in mode-locked fiber lasers. We find that when a stable soliton is generated, it's state-of-polarization shifts toward a stable state and when the soliton is generated with excess power levels it experiences relaxation oscillations in its intensity and timing. On the other hand, when a soliton is generated in an unstable state-of-polarization, it either decays in intensity until it disappears, or its temporal width decreases until it explodes into several solitons and then it disappears. We also found that when two solitons are simultaneously generated close to each other, they attract each other until they collide and merge into a single soliton. Although, these two solitons are generated with different states-of-polarization, they shift their state-of-polarization closer to each other until the polarization coincides when they collide. We support our findings by numerical calculations of a non-Lagrangian approach by simulating the Ginzburg-Landau equation governing the dynamics of solitons in a laser cavity. Our model also predicts the relaxation oscillations of stable solitons and the two types of unstable solitons observed in the experimental measurements. △ Less

Submitted 18 April, 2021; originally announced April 2021.

Comments: 10 pages, 6 figures, submitted to Opt. Exp

Intrinsically Disordered Proteins at the Nano-scale

Authors: Tamara Ehm, Hila Shinar, Sagi Meir, Amandeep Sekhon, Vaishali Sethi, Ian L. Morgan, Gil Rahamim, Omar A. Saleh, Roy Beck

Abstract: The human proteome is enriched in proteins that do not fold into a stable 3D structure. These intrinsically disordered proteins (IDPs) spontaneously fluctuate between a large number of configurations in their native form. Remarkably, the disorder does not lead to dysfunction as with denatured folded proteins. In fact, unlike denatured proteins, recent evidences strongly suggest that multiple biolo… ▽ More The human proteome is enriched in proteins that do not fold into a stable 3D structure. These intrinsically disordered proteins (IDPs) spontaneously fluctuate between a large number of configurations in their native form. Remarkably, the disorder does not lead to dysfunction as with denatured folded proteins. In fact, unlike denatured proteins, recent evidences strongly suggest that multiple biological functions stem from such structural plasticity. Here, focusing on the nanoscopic length-scale, we review the latest advances in IDP research and discuss some of the future directions in this highly promising field. △ Less

Submitted 18 January, 2021; originally announced January 2021.

Comments: 15 pages, 5 figures

Journal ref: Nano Futures 5(2), 022501 (2021)

Generating End-to-End Adversarial Examples for Malware Classifiers Using Explainability

Authors: Ishai Rosenberg, Shai Meir, Jonathan Berrebi, Ilay Gordon, Guillaume Sicard, Eli David

Abstract: In recent years, the topic of explainable machine learning (ML) has been extensively researched. Up until now, this research focused on regular ML users use-cases such as debugging a ML model. This paper takes a different posture and show that adversaries can leverage explainable ML to bypass multi-feature types malware classifiers. Previous adversarial attacks against such classifiers only add ne… ▽ More In recent years, the topic of explainable machine learning (ML) has been extensively researched. Up until now, this research focused on regular ML users use-cases such as debugging a ML model. This paper takes a different posture and show that adversaries can leverage explainable ML to bypass multi-feature types malware classifiers. Previous adversarial attacks against such classifiers only add new features and not modify existing ones to avoid harming the modified malware executable's functionality. Current attacks use a single algorithm that both selects which features to modify and modifies them blindly, treating all features the same. In this paper, we present a different approach. We split the adversarial example generation task into two parts: First we find the importance of all features for a specific sample using explainability algorithms, and then we conduct a feature-specific modification, feature-by-feature. In order to apply our attack in black-box scenarios, we introduce the concept of transferability of explainability, that is, applying explainability algorithms to different classifiers using different features subsets and trained on different datasets still result in a similar subset of important features. We conclude that explainability algorithms can be leveraged by adversaries and thus the advocates of training more interpretable classifiers should consider the trade-off of higher vulnerability of those classifiers to adversarial attacks. △ Less

Submitted 1 June, 2022; v1 submitted 28 September, 2020; originally announced September 2020.

Comments: Accepted as a conference paper at IJCNN 2020

Design and realization of a sputter deposition system for the \textit{in situ-} and \textit{in operando-}use in polarized neutron reflectometry experiments

Authors: Andreas Schmehl, Thomas Mairoser, Alexander Herrnberger, Cyril Stephanos, Stefan Meir, Benjamin Förg, Birgit Wiedemann, Peter Böni, Jochen Mannhart, Wolfgang Kreuzpaintner

Abstract: We report on the realization of a sputter deposition system for the in situ- and in operando-use in polarized neutron reflectometry experiments. Starting with the scientific requirements, which define the general design considerations, the external limitations and boundaries imposed by the available space at a neutron beamline and by the neutron and vacuum compatibility of the used materials, are… ▽ More We report on the realization of a sputter deposition system for the in situ- and in operando-use in polarized neutron reflectometry experiments. Starting with the scientific requirements, which define the general design considerations, the external limitations and boundaries imposed by the available space at a neutron beamline and by the neutron and vacuum compatibility of the used materials, are assessed. The relevant aspects are then accounted for in the realization of our highly mobile deposition system, which was designed with a focus on a quick and simple installation and removability at the beamline. Apart from the general design, the in-vacuum components, the auxiliary equipment and the remote control via a computer, as well as relevant safety aspects are presented in detail. △ Less

Submitted 19 November, 2017; originally announced November 2017.

Comments: Submitted for publication in Nuclear Inst. and Methods in Physics Research, A. (1st revised version)

Highly-Efficient Thermoelectronic Conversion of Solar Energy and Heat into Electric Power

Authors: S. Meir, C. Stephanos, T. H. Geballe, J. Mannhart

Abstract: Electric power may, in principle, be generated in a highly efficient manner from heat created by focused solar irradiation, chemical combustion, or nuclear decay by means of thermionic energy conversion. As the conversion efficiency of the thermionic process tends to be degraded by electron space charges, the efficiencies of thermionic generators have amounted to only a fraction of those fundament… ▽ More Electric power may, in principle, be generated in a highly efficient manner from heat created by focused solar irradiation, chemical combustion, or nuclear decay by means of thermionic energy conversion. As the conversion efficiency of the thermionic process tends to be degraded by electron space charges, the efficiencies of thermionic generators have amounted to only a fraction of those fundamentally possible. We show that this space-charge problem can be resolved by shaping the electric potential distribution of the converter such that the static electron space-charge clouds are transformed into an output current. Although the technical development of practical generators will require further substantial efforts, we conclude that a highly efficient transformation of heat to electric power may well be achieved. △ Less

Submitted 15 January, 2013; originally announced January 2013.

Comments: 10 pages including appendix, 4 figures

Journal ref: J. Renewable Sustainable Energy 5, 043127 (2013)