Bar-Ilan Faculty of Engineering

We design an adaptive learning algorithm for binary classification problems whose objective is to reduce the cost of misclassified instances derived from the consequences of errors. Our algorithm (Adaptive Cost-Sensitive Learning — AdaCSL) adaptively adjusts the loss function to bridge the difference between the class distributions between subgroups of samples in the training and validation data sets. This adjustment is made for samples with similar predicted probabilities, in such a way that the local cost decreases. This process usually leads to a reduction in cost when applied to the test data set (i.e., local training–test class distributions mismatch). We present empirical evidence that neural networks used with the proposed algorithm yields better cost results on several data sets compared to other approaches. In addition, the proposed AdaCSL algorithm can optimize evaluation metrics other than cost. We …

We study the problem of performing face verification with an efficient neural model . The efficiency of stems from simplifying the face verification problem from an embedding nearest neighbor search into a binary problem; each user has its own neural network . To allow information sharing between different individuals in the training set, we do not train directly but instead generate the model weights using a hypernetwork . This leads to the generation of a compact personalized model for face identification that can be deployed on edge devices. Key to the method's success is a novel way of generating hard negatives and carefully scheduling the training objectives. Our model leads to a substantially small requiring only 23k parameters and 5M floating point operations (FLOPS). We use six face verification datasets to demonstrate that our method is on par or better than state-of-the-art models, with a significantly reduced number of parameters and computational burden. Furthermore, we perform an extensive ablation study to demonstrate the importance of each element in our method.

Weak values and Kirkwood--Dirac (KD) quasiprobability distributions have been independently associated with both foundational issues in quantum theory and advantages in quantum metrology. We propose simple quantum circuits to measure weak values, KD distributions, and spectra of density matrices without the need for post-selection. This is achieved by measuring unitary-invariant, relational properties of quantum states, which are functions of Bargmann invariants, the concept that underpins our unified perspective. Our circuits also enable experimental implementation of various functions of KD distributions, such as out-of-time-ordered correlators (OTOCs) and the quantum Fisher information in post-selected parameter estimation, among others. An upshot is a unified view of nonclassicality in all those tasks. In particular, we discuss how negativity and imaginarity of Bargmann invariants relate to set coherence.

In this comprehensive review, we delve into super-resolution optical imaging techniques and their diverse applications. Our primary focus is on linear optics super-resolution methods, encompassing a wide array of concepts ranging from time multiplexing, ptychography, and deep learning-based microscopy to compressive sensing and random phase encoding techniques. Additionally, we explore compressed sensing, non-spatial resolution improvement, and sparsity-based geometric super-resolution. Furthermore, we investigate various methods based on field of view, wavelength, coherence, polarization, gray level, and code division multiplexing, as well as localization microscopy. Our review extends to stimulated emission depletion microscopy via pump-probe super-resolution techniques, providing a detailed analysis of their working applications. We then shift our attention to near-field scanning optical …

We study the problem of contextual feature selection, where the goal is to learn a predictive function while identifying subsets of informative features conditioned on specific contexts. Towards this goal, we generalize the recently proposed stochastic gates (STG) Yamada et al.[2020] by modeling the probabilistic gates as conditional Bernoulli variables whose parameters are predicted based on the contextual variables. Our new scheme, termed conditional-STG (c-STG), comprises two networks: a hypernetwork that establishes the mapping between contextual variables and probabilistic feature selection parameters and a prediction network that maps the selected feature to the response variable. Training the two networks simultaneously ensures the comprehensive incorporation of context and feature selection within a unified model. We provide a theoretical analysis to examine several properties of the proposed …

Federated learning is a technique that allows multiple entities to collaboratively train models using their data without compromising data privacy. However, despite its advantages, federated learning can be susceptible to false data injection attacks. In these scenarios, a malicious entity with control over specific agents in the network can manipulate the learning process, leading to a suboptimal model. Consequently, addressing these data injection attacks presents a significant research challenge in federated learning systems. In this paper, we propose a novel technique to detect and mitigate data injection attacks on federated learning systems. Our mitigation method is a local scheme, performed during a single instance of training by the coordinating node, allowing the mitigation during the convergence of the algorithm. Whenever an agent is suspected to be an attacker, its data will be ignored for a certain period, this decision will often be re-evaluated. We prove that with probability 1, after a finite time, all attackers will be ignored while the probability of ignoring a trustful agent becomes 0, provided that there is a majority of truthful agents. Simulations show that when the coordinating node detects and isolates all the attackers, the model recovers and converges to the truthful model.

How can non-communicating agents learn to share congested resources efficiently? This is a challenging task when the agents can access the same resource simultaneously (in contrast to multi-agent multi-armed bandit problems) and the resource valuations differ among agents. We present a fully distributed algorithm for learning to share in congested environments and prove that the agents’ regret with respect to the optimal allocation is poly-logarithmic in the time horizon. Performance in the non-asymptotic regime is illustrated in numerical simulations. The distributed algorithm has applications in cloud computing and spectrum sharing.

Optical spectroscopy the measurement of electromagnetic spectra is fundamental to various scientific domains and serves as the building block of numerous technologies. Computational spectrometry is an emerging field that employs an array of photodetectors with different spectral responses or a single photodetector device with tunable spectral response, in conjunction with numerical algorithms, for spectroscopic measurements. Compact single photodetectors made from layered materials are particularly attractive, since they eliminate the need for bulky mechanical and optical components used in traditional spectrometers and can easily be engineered as heterostructures to optimize device performance. However, compact tunable photodetectors are typically nonlinear devices and this adds complexity to extracting optical spectra from the device response. Here, we report on the training of an artificial neural network (ANN) to recover the full nonlinear spectral photoresponse of a nonlinear problem of high dimensionality of a single GeSe-InSe p-n heterojunction device. We demonstrate the functionality of a calibrated spectrometer in the spectral range of 400-1100 nm, with a small device footprint of ~25X25 micrometers, and we achieve a mean reconstruction error of 0.0002 for the power-spectrum at a spectral resolution of 0.35 nm. Using our device, we demonstrate a solution to metamerism, an apparent matching of colors with different power spectral distributions, which is a fundamental problem in optical imaging.

Intraocular pressure (IOP) measurements comprise an essential tool in modern medicine for the early diagnosis of glaucoma, the second leading cause of human blindness. The world's highest prevalence of glaucoma is in low-income countries.Current diagnostic methods require experience in running expensive equipment as well as the use of anesthetic eye drops. We present herein a remote photonic IOP biomonitoring method based on deep learning of secondary speckle patterns, captured by a fast camera, that are reflected from eye sclera stimulated by an external sound wave. By combining speckle pattern analysis with deep learning, high precision measurements are possible.

The significant heterogeneity of Wilms’ tumors between different patients is thought to arise from genetic and epigenetic distortions that occur during various stages of fetal kidney development in a way that is poorly understood. To address this, we characterized the heterogeneity of alternative mRNA splicing in Wilms’ tumors using a publicly available RNAseq dataset of high-risk Wilms’ tumors and normal kidney samples. Through Pareto task inference and cell deconvolution, we found that the tumors and normal kidney samples are organized according to progressive stages of kidney development within a triangle-shaped region in latent space, whose vertices, or “archetypes,” resemble the cap mesenchyme, the nephrogenic stroma, and epithelial tubular structures of the fetal kidney. We identified a set of genes that are alternatively spliced between tumors located in different regions of latent space and found that many of these genes are associated with the Epithelial to Mesenchymal Transition (EMT) and muscle development. Using motif enrichment analysis, we identified putative splicing regulators, some of which are associated with kidney development. Our findings provide new insights into the etiology of Wilms’ tumors and suggest that specific splicing mechanisms in early stages of development may contribute to tumor development in different patients.

The significant heterogeneity of Wilms’ tumors between different patients is thought to arise from genetic and epigenetic distortions that occur during various stages of fetal kidney development in a way that is poorly understood. To address this, we characterized the heterogeneity of alternative mRNA splicing in Wilms’ tumors using a publicly available RNAseq dataset of high-risk Wilms’ tumors and normal kidney samples. Through Pareto task inference and cell deconvolution, we found that the tumors and normal kidney samples are organized according to progressive stages of kidney development within a triangle-shaped region in latent space, whose vertices, or “archetypes,” resemble the cap mesenchyme, the nephrogenic stroma, and epithelial tubular structures of the fetal kidney. We identified a set of genes that are alternatively spliced between tumors located in different regions of latent space and found that many of these genes are associated with the Epithelial to Mesenchymal Transition (EMT) and muscle development. Using motif enrichment analysis, we identified putative splicing regulators, some of which are associated with kidney development. Our findings provide new insights into the etiology of Wilms’ tumors and suggest that specific splicing mechanisms in early stages of development may contribute to tumor development in different patients.

Anomalies (or outliers) are prevalent in real-world empirical observations and potentially mask important underlying structures. Accurate identification of anomalous samples is crucial for the success of downstream data analysis tasks. To automatically identify anomalies, we propose Probabilistic Robust AutoEncoder (PRAE). PRAE aims to simultaneously remove outliers and identify a low-dimensional representation for the inlier samples. We first present the Robust AutoEncoder (RAE) objective as a minimization problem for splitting the data into inliers and outliers. Our objective is designed to exclude outliers while including a subset of samples (inliers) that can be effectively reconstructed using an AutoEncoder (AE). RAE minimizes the autoencoder's reconstruction error while incorporating as many samples as possible. This could be formulated via regularization by subtracting an norm counting the number of selected samples from the reconstruction term. Unfortunately, this leads to an intractable combinatorial problem. Therefore, we propose two probabilistic relaxations of RAE, which are differentiable and alleviate the need for a combinatorial search. We prove that the solution to the PRAE problem is equivalent to the solution of RAE. We use synthetic data to show that PRAE can accurately remove outliers in a wide range of contamination levels. Finally, we demonstrate that using PRAE for anomaly detection leads to state-of-the-art results on various benchmark datasets.

We consider the weighted completion time minimization problem for capacitated parallel machines, which is a fundamental problem in modern cloud computing environments. In our setting, the processed jobs may be of varying duration, require different resources, and be of unequal importance (weight). Each server (machine) can process multiple concurrent jobs up to its capacity. We study heuristic approaches with provable approximation guarantees and offer an algorithm that prioritizes the jobs with the smallest volume-by-weight ratio. We bound the algorithm’s approximation ratio using a decreasing function of the ratio between the highest resource demand of any job and the server’s capacity. Thereafter, we create a hybrid, constant approximation algorithm for two or more machines. We also develop a constant approximation algorithm for the case of a single machine. Via a numerical study and a mixed …

A critical current challenge in the development of all-solid-state lithium batteries (ASSLBs) is reducing the cost of fabrication without compromising the performance. Here we report a sulfide ASSLB based on a high-energy, Co-free LiNiO2 cathode with a robust outside-in structure. This promising cathode is enabled by the high-pressure O2 synthesis and subsequent atomic layer deposition of a unique ultrathin LixAlyZnzOδ protective layer comprising a LixAlyZnzOδ surface coating region and an Al and Zn near-surface doping region. This high-quality artificial interphase enhances the structural stability and interfacial dynamics of the cathode as it mitigates the contact loss and continuous side reactions at the cathode/solid electrolyte interface. As a result, our ASSLBs exhibit a high areal capacity (4.65 mAh cm−2), a high specific cathode capacity (203 mAh g−1), superior cycling stability (92% capacity retention …

Immunoglobulins (IGs), critical components of the human immune system, are composed of heavy and light protein chains encoded at three genomic loci. The IG Kappa (IGK) chain locus consists of two large, inverted segmental duplications. The complexity of IG loci has hindered effective use of standard high-throughput methods for characterizing genetic variation within these regions. To overcome these limitations, we leverage long-read sequencing to create haplotype-resolved IGK assemblies in an ancestrally diverse cohort (n=36), representing the first comprehensive description of IGK haplotype variation at population-scale. We identify extensive locus polymorphism, including novel single nucleotide variants (SNVs) and a common novel ~24.7 Kbp structural variant harboring a functional IGKV gene. Among 47 functional IGKV genes, we identify 141 alleles, 64 (45.4%) of which were not previously curated. We report inter-population differences in allele frequencies for 14 of the IGKV genes, including alleles unique to specific populations within this dataset. Finally, we identify haplotypes carrying signatures of gene conversion that associate with enrichment of SNVs in the IGK distal region. These data provide a critical resource of curated genomic reference information from diverse ancestries, laying a foundation for advancing our understanding of population-level genetic variation in the IGK locus.

Digital holographic microscopy (DHM) is a very popular interferometric technique for quantitative phase imaging (QPI). In DHM, an interferometer is combined with a microscope to create interference between an imaging beam containing information about the analysed sample and a clear reference beam carrying no sample information. To exploit the capability of reference beam in terms of useful sample information, we have recently proposed Cepstrum-based Interferometric Microscopy (CIM) [Opt. Las. Tech. 174, 110,626 (2024)] as a novel methodology involving the interference of two imaging beams carrying different sample information and to accurately retrieve quantitative phase data of both beams. In the earlier implementation, proof-of-concept of CIM was demonstrated for a Michelson-based layout requiring manual adjustments during the CIM methodology and validated only for low numerical aperture (NA …

This letter studies the Friedkin-Johnsen (FJ) model with diminishing competition, or stubbornness. The original FJ model assumes fixed competition that is manifested through a constant weight that each agent gives to its initial opinion in addition to its contribution through a consensus dynamic. This letter investigates the effect of diminishing competition on the convergence point and speed of the FJ dynamics. We show that, if the competition is uniform across agents and vanishes asymptotically, the convergence point coincides with the nominal consensus reached with no competition. However, the diminishing competition slows down convergence according to its own rate of decay. We evaluate this phenomenon analytically and provide upper and lower bounds on the convergence rate. If competition is not uniform across clients, we show that the convergence point may not coincide with the nominal consensus point. Finally, we evaluate and validate our analytical insights numerically.

This paper presents a Multi-modal Emotion Recognition (MER) system designed to enhance emotion recognition accuracy in challenging acoustic conditions. Our approach combines a modified and extended Hierarchical Token-semantic Audio Transformer (HTS-AT) for multi-channel audio processing with an R(2+1)D Convolutional Neural Networks (CNN) model for video analysis. We evaluate our proposed method on a reverberated version of the Ryerson audio-visual database of emotional speech and song (RAVDESS) dataset using synthetic and real-world Room Impulse Responsess (RIRs). Our results demonstrate that integrating audio and video modalities yields superior performance compared to uni-modal approaches, especially in challenging acoustic conditions. Moreover, we show that the multimodal (audiovisual) approach that utilizes multiple microphones outperforms its single-microphone counterpart.

Quantum query complexity mainly studies the number of queries needed to learn some property of a black box with high probability. A closely related question is how well an algorithm can succeed with this learning task using only a fixed number of queries. In this work, we propose measuring an algorithm's performance using the mutual information between the output and the actual value. A key observation is that if an algorithm is only allowed to make a single query and the goal is to optimize this mutual information, then we obtain a task which is similar to a basic task of quantum communication, where one attempts to maximize the mutual information of the sender and receiver. We make this analogy precise by formally considering the oracle as a separate subsystem, whose state records the unknown oracle identity. The oracle query prepares a state, which is then measured; and the target property of the oracle plays the role of a message that should be deduced from the measurement outcome. Thus we obtain a link between the optimal single-query algorithm and minimization of the extent of quantum correlations between the oracle and the computer subsystems. We also find a lower bound on this mutual information, which is related to quantum coherence. These results extend to multiple-query non-adaptive algorithms. As a result, we gain insight into the task of finding the optimal non-adaptive algorithm that uses at most a fixed number of queries, for any oracle problem.

Modulation of electronic properties of materials by electric fields is central to the operation of modern semiconductor devices, providing access to complex electronic behaviors and greater freedom in tuning the energy bands of materials. Here, we explore one-dimensional superlattices induced by a confining electrostatic potential in monolayer MoS, a prototypical two-dimensional semiconductor. Using first-principles calculations, we show that periodic potentials applied to monolayer MoS induce electrostatic superlattices in which the response is dominated by structural distortions relative to purely electronic effects. These structural distortions reduce the intrinsic band gap of the monolayer substantially while also polarizing the monolayer through piezoelectric coupling, resulting in spatial separation of charge carriers as well as Stark shifts that produce dispersive minibands. Importantly, these minibands inherit the valley-selective magnetic properties of monolayer MoS, enabling fine control over spin-valley coupling in MoS and similar transition-metal dichalcogenides.

This paper presents a scalable, fully-unrolled Number Theory Transform (NTT) architecture that eliminates memory registers. This architecture leverages Dual-Mode Logic (DML) technology to achieve configurability for various input sizes ( N ) and bit-widths ( K ). The design targets hardware acceleration, aiming to deliver NTT functionality at the speed of signal propagation without relying on registers or incurring pipeline-related trade-offs. Our implementation offers several potential optimizations, including a unique, fully-combinational, and low-cost modular reduction technique within the K-RED algorithm. Furthermore, it eliminates memory access penalties and control overhead. The widespread adoption of NTT and similar repeated structures across various fields underscores the broad applicability of our findings. Similar to large-scale DML-based architectures (e.g., MAC multipliers), our DML-NTT design can …