Bar-Ilan Faculty of Engineering

Temporal alignment of sequences is a fundamental challenge in many applications, such as computer vision and bioinformatics, where local time shifting needs to be accounted for. Misalignment can lead to poor model generalization, especially in high-dimensional sequences. Existing methods often struggle with optimization when dealing with high-dimensional sparse data, falling into poor alignments. Feature selection is frequently used to enhance model performance for sparse data. However, a fixed set of selected features would not generally work for dynamically changing sequences and would need to be modified based on the state of the sequence. Therefore, modifying the selected feature based on contextual input would result in better alignment. Our suggested method, Conditional Deep Canonical Temporal Time Warping (CDCTW), is designed for temporal alignment in sparse temporal data to address these challenges. CDCTW enhances alignment accuracy for high dimensional time-dependent views be performing dynamic time warping on data embedded in maximally correlated subspace which handles sparsity with novel feature selection method. We validate the effectiveness of CDCTW through extensive experiments on various datasets, demonstrating superior performance over previous techniques.

Enhancing resilience in distributed networks in the face of malicious agents is an important problem for which many key theoretical results and applications require further development and characterization. This work focuses on the problem of distributed optimization in multi-agent cyberphysical systems, where a legitimate agent's dynamic is influenced both by the values it receives from potentially malicious neighboring agents, and by its own self-serving target function. We develop a new algorithmic and analytical framework to achieve resilience for the class of problems where stochastic values of trust between agents exist and can be exploited. In this case we show that convergence to the true global optimal point can be recovered, both in mean and almost surely, even in the presence of malicious agents. Furthermore, we provide expected convergence rate guarantees in the form of upper bounds on the expected squared distance to the optimal value. Finally, we present numerical results that validate the analytical convergence guarantees we present in this paper even when the malicious agents compose the majority of agents in the network.

Training and fine-tuning large language models (LLMs) come with challenges related to memory and computational requirements due to the increasing size of the model weights and the optimizer states. Various techniques have been developed to tackle these challenges, such as low-rank adaptation (LoRA), which involves introducing a parallel trainable low-rank matrix to the fixed pre-trained weights at each layer. However, these methods often fall short compared to the full-rank weight training approach, as they restrict the parameter search to a low-rank subspace. This limitation can disrupt training dynamics and require a full-rank warm start to mitigate the impact. In this paper, we introduce a new method inspired by a phenomenon we formally prove: as training progresses, the rank of the estimated layer gradients gradually decreases, and asymptotically approaches rank one. Leveraging this, our approach involves adaptively reducing the rank of the gradients during Adam optimization steps, using an efficient online-updating low-rank projections rule. We further present a randomized SVD scheme for efficiently finding the projection matrix. Our technique enables full-parameter fine-tuning with adaptive low-rank gradient updates, significantly reducing overall memory requirements during training compared to state-of-the-art methods while improving model performance in both pretraining and fine-tuning. Finally, we provide a convergence analysis of our method and demonstrate its merits for training and fine-tuning language and biological foundation models.

We study the question of whether a sequence of positive integers is the degree sequence of some outerplanar (a.k.a. 1-page book embeddable) graph G. If so, G is an outerplanar realization of d and d is an outerplanaric sequence. The case where is easy, as d has a realization by a forest (which is trivially an outerplanar graph). In this paper, we consider the family of all sequences d of even sum , where is the number of x’s in d. (The second inequality is a necessary condition for a sequence d with to be outerplanaric.) We partition into two disjoint subfamilies, , such that every sequence in is provably non-outerplanaric, and every sequence in is given a realizing graph G enjoying a 2-page book embedding (and moreover, one of the pages is also bipartite).

Leakage-resilient cryptography aims to protect cryptographic primitives from so-called “side channel attacks” that exploit their physical implementation to learn their input or secret state. Starting from the works of Ishai, Sahai and Wagner (CRYPTO ‘03) and Micali and Reyzin (TCC ‘04), most works on leakage-resilient cryptography either focus on protecting general computations, such as circuits or multiparty computation protocols, or on specific non-interactive primitives such as storage, encryption, and signatures. This work focuses on leakage resilience for the middle ground, namely for distributed and interactive cryptographic primitives. Our main technical contribution is designing the first secret sharing scheme that is equivocal, resists adaptive probing of a constant fraction of bits from each share, while incurs only a constant blowup in share size. Equivocation is a strong leakage-resilience guarantee, recently …

We study the question of whether a sequence d=(d 1,…, d n) of positive integers is the degree sequence of some outerplanar graph G. If so, G is an outerplanar realization of d and d is an outerplanaric sequence. The case where∑ d≤ 2 n− 2 is easy, as d has a realization by a forest. In this paper, we consider the family D of all sequences d of even sum 2 n≤∑ d≤ 4 n− 6− 2 ω 1, where ω x is the number of x's in d. We partition D into two disjoint subfamilies, D= D N O P∪ D 2 P B E, such that every sequence in D N O P is provably non-outerplanaric, and every sequence in D 2 P B E is given a realizing graph G enjoying a 2-page book embedding (and moreover, one of the pages is also bipartite).

Machine learning (ML) has shown great potential in the adaptive immune receptor repertoire (AIRR) field. However, there is a lack of large-scale ground-truth experimental AIRR data suitable for AIRR-ML-based disease diagnostics and therapeutics discovery. Simulated ground-truth AIRR data are required to complement the development and benchmarking of robust and interpretable AIRR-ML methods where experimental data is currently inaccessible or insufficient. The challenge for simulated data to be useful is incorporating key features observed in experimental repertoires. These features, such as antigen or disease-associated immune information, cause AIRR-ML problems to be challenging. Here, we introduce LIgO, a software suite, which simulates AIRR data for the development and benchmarking of AIRR-ML methods. LIgO incorporates different types of immune information both on the receptor and …

This paper introduces a multi-microphone method for extracting a desired speaker from a mixture involving multiple speakers and directional noise in a reverberant environment. In this work, we propose leveraging the instantaneous relative transfer function (RTF), estimated from a reference utterance recorded in the same position as the desired source. The effectiveness of the RTF-based spatial cue is compared with direction of arrival (DOA)-based spatial cue and the conventional spectral embedding. Experimental results in challenging acoustic scenarios demonstrate that using spatial cues yields better performance than the spectral-based cue and that the instantaneous RTF outperforms the DOA-based spatial cue.

The scattering-type Scanning Near-Field Optical Microscope (s-SNOM) is acknowledged as an excellent tool to investigate the optical properties of different materials and biological samples at the nanoscale. In this study we show that s-SNOM data are susceptible to being affected by specific artefacts related to the light diffraction phenomena and to stray contributions from shallow buried, contrast-active, structures. We focus on discussing the diffraction contributions from sample edges, next to those corresponding to one- or two-dimensional periodic structures, and undesired contributions from shallow buried periodic features. Each scenario was examined individually through both experimental methods and simulations. Our experimental findings reveal that such artefacts affect not only s-SNOM images demodulated at the direct-current (DC) component and the fundamental frequency, but also images demodulated at higher harmonic frequencies. We show that image artefacts caused by diffraction resemble the undesirable effects caused by illumination with a laser beam of unstable intensity, and that buried features can yield s-SNOM signals that cannot be distinguished from those originating from the sample surface, in absence of prior knowledge of the sample structure. Performed simulations confirm these experimental findings. This work enhances the understanding of s-SNOM data and paves the way for new data acquisition and postprocessing methods that can enable next-generation s-SNOM imaging and spectroscopy with significantly enhanced signal-to-noise ratio and resolution.

Fusing information from different modalities can enhance data analysis tasks, including clustering. However, existing multi-view clustering (MVC) solutions are limited to specific domains or rely on a suboptimal and computationally demanding two-stage procedure of representation and clustering. We propose an end-to-end deep learning-based MVC framework for general data (image, tabular, etc.). Our approach involves learning meaningful fused data representations with a novel permutation-based canonical correlation objective. Concurrently, we learn cluster assignments by identifying consistent pseudo-labels across multiple views. We demonstrate the effectiveness of our model using ten MVC benchmark datasets. Theoretically, we show that our model approximates the supervised linear discrimination analysis (LDA) representation. Additionally, we provide an error bound induced by false-pseudo label annotations.

This paper presents an extension of time multiplexing super-resolution imaging concept to allow imaging through scattering medium. The presented concept includes laser illuminating an object through a diffuser. The technique performs time multiplexing super-resolved imaging through this diffuser while using the unknown speckle pattern the diffuser projects on the imaged object to enhance the resolution at which the object is imaged. Thus, unlike in conventional case where imaging through a diffuser or other scattering medium destroys the imaging resolution, here the speckle pattern the diffuser generates assists in performing super-resolved imaging. Experimental results demonstrate the efficacy of the approach, showing significant improvement in image quality and resolution.

Arterial oxygen saturation (SpO2), a key indicator of respiratory health, reflects the proportion of oxygenated hemoglobin in the blood and is essential for monitoring conditions such as hypoxia. Traditional pulse oximetry methods use multiple wavelengths to calculate SpO2, which cause errors due to differences in optical pathlengths, affected by different scattering coefficients. This study presents an optical biosensor for non-invasive measurement of SpO2, utilizing the iso-pathlength (IPL) point concept. Our biosensor overcomes the inherent limitations of the classic method by using a single light source and detecting reemitted light at the IPL point, where light intensity is invariant to scattering. This enables accurate SpO2 measurements without the need for external calibration. The biosensor operates with a red LED at 655nm and five photodetectors, one positioned at the IPL point, which allows the extraction of …

High-performance water-based inkjet inks are critical for advancing inkjet printing technology. The performance of water-based inkjet inks depends largely on the dispersion stability of organic pigments. This imposes higher demands on the performance of polymeric dispersants. However, the relatively weak interaction between polymeric dispersants and organic pigments limits their performance in water-based inkjet inks. Consequently, it is crucial to seek dispersants that exhibit stronger interactions with pigments, alongside high performance, and universality. In this work, five types of polymeric nanoparticles (PNPs) with anion-π groups were synthesized via a simple emulsion polymerization method. Compared to traditional polymeric dispersants, anion-π type PNPs exhibited significant advantages including low viscosity, solvent resistance, and high temperature resistance. Stronger interactions, including salt-bridge hydrogen bondings (H-bonds) and π–π interactions, between these PNPs and different types of organic pigments were demonstrated by FTIR, UV-Vis, and XPS spectral tests. In particular, PNPs-5, bearing -PhSO3− groups, exhibited the strongest interaction with the organic pigments. The water-based inkjet inks, formulated with PNPs-5 serving as a dispersant, exhibited remarkable dispersion stability and outstanding weatherability. This work rationally constructs a strategy for preparing universally applicable polymeric dispersants to enhance the dispersion of pigments in water-based inkjet inks, thereby presenting a broader perspective for applications in the field of inkjet printing.

Photoluminescence of non-aromatic supramolecular chemical assemblies has attracted considerable attention in recent years due to its potential for use in molecular sensing and imaging technologies. The underlying structural origins, the mechanisms of light emission in these systems, and the generality of this phenomenon remain elusive. Here, we demonstrate that crystals of L-Cysteine (Cys) formed in heavy water () exhibit distinct packing and hydrogen-bond networks, resulting in significantly enhanced photoluminescence compared to those prepared in . Using advanced excited-state simulations, we elucidate the nature of electronic transitions that activate vibrational modes of Cys in , particularly those involving thiol (S-H) and amine (C-N) groups, which lead to non-radiative decay. For the crystal formed in , these modes appear to be more constrained, and we also observe intersystem crossing from the singlet to the triplet state, indicating a potentially more complex light emission mechanism. Our findings provide new insights into this intriguing phenomenon and introduce innovative design principles for generating emergent fluorophores.

In image processing, solving inverse problems is the task of finding plausible reconstructions of an image that was corrupted by some (usually known) degradation model. Commonly, this process is done using a generative image model that can guide the reconstruction towards solutions that appear natural. The success of diffusion models over the last few years has made them a leading candidate for this task. However, the sequential nature of diffusion models makes this conditional sampling process challenging. Furthermore, since diffusion models are often defined in the latent space of an autoencoder, the encoder-decoder transformations introduce additional difficulties. Here, we suggest a novel sampling method based on sequential Monte Carlo (SMC) in the latent space of diffusion models. We use the forward process of the diffusion model to add additional auxiliary observations and then perform an SMC sampling as part of the backward process. Empirical evaluations on ImageNet and FFHQ show the benefits of our approach over competing methods on various inverse problem tasks.

This work introduces a novel deep-learning approach for estimating age from a single facial image by refining an initial age estimate. The refinement leverages a reference face database of individuals with similar ages and appearances. We employ a network that estimates age differences between an input image and reference images with known ages, thus refining the initial estimate. Our method explicitly models age-dependent facial variations using differential regression, yielding improved accuracy compared to conventional absolute age estimation. Additionally, we introduce an age augmentation scheme that iteratively refines initial age estimates by modeling their error distribution during training. This iterative approach further enhances the initial estimates. Our approach surpasses existing methods, achieving state-of-the-art accuracy on the MORPH II and CACD datasets. Furthermore, we examine the biases inherent in contemporary state-of-the-art age estimation techniques.

The inference of gene regulatory networks (GRNs) from single-cell RNAseq data allows for mechanistic characterization of the different cell states and their dynamics in complex biological processes. While numerous algorithms have been proposed to infer GRNs from single-cell transcriptomic data, multiple network solutions may explain the same dataset, posing a challenge for biologically meaningful interpretation. Here, we use the Reasoning Engine for Interaction Networks (RE:IN), a computational tool based on formal reasoning, to characterize GRN ensembles in the context of acute kidney injury (AKI). To this end, we applied RE:IN to a single-cell RNAseq dataset from a mouse ischemia reperfusion injury (IRI) model, focusing on distinct proximal tubule cell states related to kidney injury and repair. We first created an Abstract Boolean Network (ABN) model for the kidney using RE:IN and synthesized an …

Rapid, sensitive, and high-throughput detection of biomarkers at low concentrations is crucial for early disease diagnosis. Many sensitive immunoassays use magnetic beads to capture fluorescently labeled targets, but quantifying these targets involves detecting the fluorescent signal from individual beads, which is time-consuming and requires a costly detection system. Additionally, there is often a trade-off between sensitivity, speed, throughput, and ease of use. A new technology, high-throughput optical modulation biosensing (OMB), enables reading a 96-well plate within 10 minutes. In OMB, a cylindrical permanent magnet immobilizes the magnetic beads at the illumination spot. Then, a laser beam is manipulated between the magnetic beads cluster and the background solution, effectively subtracting noise and reducing the need for washing and separation steps, which are usually incorporated in heterogeneous assays. This technology has evolved into a fully automated platform with high sensitivity and throughput, allowing much faster turnaround time and better sensitivity than the state-of-the-art methods, like enzyme-linked immunosorbent assay (ELISA)(for protein detection) and real-time PCR (for RNA/DNA detection). Here, we provide a comprehensive review of this technology, its development, and its applications in rapid, highly sensitive detection of proteins (eg, human Interleukin-8) and viruses (eg, SARS-CoV-2).

Conformal Prediction (CP) is a method to control prediction uncertainty by producing a small prediction set, ensuring a predetermined probability that the true class lies within this set. This is commonly done by defining a score, based on the model predictions, and setting a threshold on this score using a validation set. In this study, we address the problem of CP calibration when we only have access to a validation set with noisy labels. We show how we can estimate the noise-free conformal threshold based on the noisy labeled data. Our solution is flexible and can accommodate various modeling assumptions regarding the label contamination process, without needing any information about the underlying data distribution or the internal mechanisms of the machine learning classifier. We develop a coverage guarantee for uniform noise that is effective even in tasks with a large number of classes. We dub our approach Noise-Aware Conformal Prediction (NACP) and show on several natural and medical image classification datasets, including ImageNet, that it significantly outperforms current noisy label methods and achieves results comparable to those obtained with a clean validation set.

We present the first algorithm for testing equivalence between two continuous distributions using differential privacy (DP). Our algorithm is a private version of the algorithm of Diakonikolas et al. The algorithm of Diakonikolas et al uses the data itself to repeatedly discretize the real line so that---when the two distributions are far apart in -norm---one of the discretized distributions exhibits large -norm difference; and upon repeated sampling such large gap would be detected. Designing its private analogue poses two difficulties. First, our DP algorithm can not resample new datapoints as a change to a single datapoint may lead to a very large change in the descretization of the real line. In contrast, the (sorted) index of the discretization point changes only by between neighboring instances, and so we use a novel algorithm that set the discretization points using random Bernoulli noise, resulting in only a few buckets being affected under the right coupling. Second, our algorithm, which doesn't resample data, requires we also revisit the utility analysis of the original algorithm and prove its correctness wrt the original sorted data; a problem we tackle using sampling a subset of Poisson-drawn size from each discretized bin. Lastly, since any distribution can be reduced to a continuous distribution, our algorithm is successfully carried to multiple other families of distributions and thus has numerous applications.

Individuals have the remarkable ability to differentiate between speakers and focus on a particular speaker, even amidst complex acoustic environments with multiple speakers, background noise and reverberations. This selective auditory attention, often illustrated by the cocktail party problem, has been extensively researched. With a considerable portion of the population experiencing hearing impairment and requiring hearing aids, there arises a necessity to separate and decode auditory signals artificially. The linearly constrained minimum variance (LCMV) beamforming design criterion has proven effective in isolating the desired source by steering a beam toward the target speaker while creating a null toward the interfering source. Preserving the binaural cues, e.g., interaural time difference (ITFD) and interaural level difference (ILD), is a prerequisite for producing a beamformer output suitable for hearing aid applications. For that, the binaural linearly constrained minimum variance (BLCMV) beamformer generates two outputs that satisfy the standard LCMV criterion while preserving the binaural cues between the left-ear and right-ear outputs. Identifying the attended speaker from the separated speakers and distinguishing it from the unattended speaker poses a fundamental challenge in the beamformer design. Several studies showed the ability to encode essential features of the attended speech from the cortex neural response, as recorded by the electroencephalography (EEG) signals. This led to the development of several algorithms addressing the auditory attention decoder (AAD) task. This paper investigates two neural network …