Israel Cohen

Eyal Hanania, Ilya Volovik, Lilach Barkat et al.

T1 mapping is a quantitative magnetic resonance imaging (qMRI) technique that has emerged as a valuable tool in the diagnosis of diffuse myocardial diseases. However, prevailing approaches have relied heavily on breath-hold sequences to eliminate respiratory motion artifacts. This limitation hinders accessibility and effectiveness for patients who cannot tolerate breath-holding. Image registration can be used to enable free-breathing T1 mapping. Yet, inherent intensity differences between the different time points make the registration task challenging. We introduce PCMC-T1, a physically-constrained deep-learning model for motion correction in free-breathing T1 mapping. We incorporate the signal decay model into the network architecture to encourage physically-plausible deformations along the longitudinal relaxation axis. We compared PCMC-T1 to baseline deep-learning-based image registration approaches using a 5-fold experimental setup on a publicly available dataset of 210 patients. PCMC-T1 demonstrated superior model fitting quality (R2: 0.955) and achieved the highest clinical impact (clinical score: 3.93) compared to baseline methods (0.941, 0.946 and 3.34, 3.62 respectively). Anatomical alignment results were comparable (Dice score: 0.9835 vs. 0.984, 0.988). Our code and trained models are available at https://github.com/eyalhana/PCMC-T1.

Daniel Ohayon, Itay Lamprecht, Itay Hubara et al.

Modern large language models increasingly require long contexts for reasoning and multi-document tasks, but attention's quadratic complexity creates a severe computational bottleneck. We present Block-Sparse FlashAttention (BSFA), a drop-in replacement that accelerates long-context inference while preserving model quality. Unlike methods that predict importance before computing scores, BSFA computes exact query-key similarities to select the top-k most important value blocks for each query. By comparing per-block maximum scores against calibrated thresholds, we skip approximately 50% of the computation and memory transfers for pruned blocks. Our training-free approach requires only a one-time threshold calibration on a small dataset to learn the per-layer and per-head attention score distributions. We provide a CUDA kernel implementation that can be used as a drop-in replacement for FlashAttention. On Llama-3.1-8B, BSFA achieves up to 1.10x speedup on real-world reasoning benchmarks and up to 1.24x for needle-in-a-haystack retrieval tasks while maintaining above 99% baseline accuracy, with certain configurations even improving accuracy by focusing on the most relevant content, substantially outperforming existing sparse attention methods. The implementation is available at https://github.com/Danielohayon/Block-Sparse-Flash-Attention

Gregory Ciccarelli, Jarred Barber, Arun Nair et al.

We review current solutions and technical challenges for automatic speech recognition, keyword spotting, device arbitration, speech enhancement, and source localization in multidevice home environments to provide context for the INTERSPEECH 2022 special session, "Challenges and opportunities for signal processing and machine learning for multiple smart devices". We also identify the datasets needed to support these research areas. Based on the review and our research experience in the multi-device domain, we conclude with an outlook on the future evolution

Eyal Hanania, Adi Zehavi-Lenz, Ilya Volovik et al.

Cardiac T1 mapping is a valuable quantitative MRI technique for diagnosing diffuse myocardial diseases. Traditional methods, relying on breath-hold sequences and cardiac triggering based on an ECG signal, face challenges with patient compliance, limiting their effectiveness. Image registration can enable motion-robust cardiac T1 mapping, but inherent intensity differences between time points pose a challenge. We present MBSS-T1, a subject-specific self-supervised model for motion correction in cardiac T1 mapping. Physical constraints, implemented through a loss function comparing synthesized and motion-corrected images, enforce signal decay behavior, while anatomical constraints, applied via a Dice loss, ensure realistic deformations. The unique combination of these constraints results in motion-robust cardiac T1 mapping along the longitudinal relaxation axis. In a 5-fold experiment on a public dataset of 210 patients (STONE sequence) and an internal dataset of 19 patients (MOLLI sequence), MBSS-T1 outperformed baseline deep-learning registration methods. It achieved superior model fitting quality ($R^2$: 0.975 vs. 0.941, 0.946 for STONE; 0.987 vs. 0.982, 0.965 for MOLLI free-breathing; 0.994 vs. 0.993, 0.991 for MOLLI breath-hold), anatomical alignment (Dice: 0.89 vs. 0.84, 0.88 for STONE; 0.963 vs. 0.919, 0.851 for MOLLI free-breathing; 0.954 vs. 0.924, 0.871 for MOLLI breath-hold), and visual quality (4.33 vs. 3.38, 3.66 for STONE; 4.1 vs. 3.5, 3.28 for MOLLI free-breathing; 3.79 vs. 3.15, 2.84 for MOLLI breath-hold). MBSS-T1 enables motion-robust T1 mapping for broader patient populations, overcoming challenges such as suboptimal compliance, and facilitates free-breathing cardiac T1 mapping without requiring large annotated datasets. Our code is available at https://github.com/TechnionComputationalMRILab/MBSS-T1.

Luoxiao Yang, Yun Wang, Xinqi Fan et al.

Time series forecasting (TSF) possesses great practical values in various fields, including power and energy, transportation, etc. TSF methods have been studied based on knowledge from classical statistics to modern deep learning. Yet, all of them were developed based on one fundamental concept, the numerical data fitting. Thus, the models developed have long been known to be problem-specific and lacking application generalizability. Practitioners expect a TSF foundation model that serves TSF tasks in different applications. The central question is then how to develop such a TSF foundation model. This paper offers one pioneering study in the TSF foundation model development method and proposes a vision intelligence-powered framework, ViTime, for the first time. ViTime fundamentally shifts TSF from numerical fitting to operations based on a binary image-based time series metric space and naturally supports both point and probabilistic forecasting. We also provide rigorous theoretical analyses of ViTime, including quantization-induced system error bounds and principled strategies for optimal parameter selection. Furthermore, we propose RealTS, an innovative synthesis algorithm generating diverse and realistic training samples, effectively enriching the training data and significantly enhancing model generalizability. Extensive experiments demonstrate ViTime's state-of-the-art performance. In zero-shot scenarios, ViTime outperforms TimesFM by 9-15\%. With just 10\% fine-tuning data, ViTime surpasses both leading foundation models and fully-supervised benchmarks, a gap that widens with 100\% fine-tuning. ViTime also exhibits exceptional robustness, effectively handling missing data and outperforming TimesFM by 20-30\% under various data perturbations, validating the power of its visual space data operation paradigm.

Gil Weissman, Amir Ivry, Israel Cohen

The proliferation of satellite constellations, coupled with reduced tasking latency and diverse sensor capabilities, has expanded the opportunities for automated Earth observation. This paper introduces a fully automated Tip-and-Cue framework designed for satellite imaging tasking and scheduling. In this context, tips are generated from external data sources or analyses of prior satellite imagery, identifying spatiotemporal targets and prioritizing them for downstream planning. Corresponding cues are the imaging tasks formulated in response, which incorporate sensor constraints, timing requirements, and utility functions. The system autonomously generates candidate tasks, optimizes their scheduling across multiple satellites using continuous utility functions that reflect the expected value of each observation, and processes the resulting imagery using artificial-intelligence-based models, including object detectors and vision-language models. Structured visual reports are generated to support both interpretability and the identification of new insights for downstream tasking. The efficacy of the framework is demonstrated through a maritime vessel tracking scenario, utilizing Automatic Identification System (AIS) data for trajectory prediction, targeted observations, and the generation of actionable outputs. Maritime vessel tracking is a widely researched application, often used to benchmark novel approaches to satellite tasking, forecasting, and analysis. The system is extensible to broader applications such as smart-city monitoring and disaster response, where timely tasking and automated analysis are critical.

Yehoshua Dissen, Shiry Yonash, Israel Cohen et al.

In the realm of automatic speech recognition (ASR), robustness in noisy environments remains a significant challenge. Recent ASR models, such as Whisper, have shown promise, but their efficacy in noisy conditions can be further enhanced. This study is focused on recovering from packet loss to improve the word error rate (WER) of ASR models. We propose using a front-end adaptation network connected to a frozen ASR model. The adaptation network is trained to modify the corrupted input spectrum by minimizing the criteria of the ASR model in addition to an enhancement loss function. Our experiments demonstrate that the adaptation network, trained on Whisper's criteria, notably reduces word error rates across domains and languages in packet-loss scenarios. This improvement is achieved with minimal affect to Whisper model's foundational performance, underscoring our method's practicality and potential in enhancing ASR models in challenging acoustic environments.

Amir Ivry, Israel Cohen, Baruch Berdugo

Human subjective evaluation is optimal to assess speech quality for human perception. The recently introduced deep noise suppression mean opinion score (DNSMOS) metric was shown to estimate human ratings with great accuracy. The signal-to-distortion ratio (SDR) metric is widely used to evaluate residual-echo suppression (RES) systems by estimating speech quality during double-talk. However, since the SDR is affected by both speech distortion and residual-echo presence, it does not correlate well with human ratings according to the DNSMOS. To address that, we introduce two objective metrics to separately quantify the desired-speech maintained level (DSML) and residual-echo suppression level (RESL) during double-talk. These metrics are evaluated using a deep learning-based RES-system with a tunable design parameter. Using 280 hours of real and simulated recordings, we show that the DSML and RESL correlate well with the DNSMOS with high generalization to various setups. Also, we empirically investigate the relation between tuning the RES-system design parameter and the DSML-RESL tradeoff it creates and offer a practical design scheme for dynamic system requirements.

Deborah Pereg, Israel Cohen, Anthony A. Vassiliou

In sparse coding, we attempt to extract features of input vectors, assuming that the data is inherently structured as a sparse superposition of basic building blocks. Similarly, neural networks perform a given task by learning features of the training data set. Recently both data-driven and model-driven feature extracting methods have become extremely popular and have achieved remarkable results. Nevertheless, practical implementations are often too slow to be employed in real-life scenarios, especially for real-time applications. We propose a speed-up upgraded version of the classic iterative thresholding algorithm, that produces a good approximation of the convolutional sparse code within 2-5 iterations. The speed advantage is gained mostly from the observation that most solvers are slowed down by inefficient global thresholding. The main idea is to normalize each data point by the local receptive field energy, before applying a threshold. This way, the natural inclination towards strong feature expressions is suppressed, so that one can rely on a global threshold that can be easily approximated, or learned during training. The proposed algorithm can be employed with a known predetermined dictionary, or with a trained dictionary. The trained version is implemented as a neural net designed as the unfolding of the proposed solver. The performance of the proposed solution is demonstrated via the seismic inversion problem in both synthetic and real data scenarios. We also provide theoretical guarantees for a stable support recovery. Namely, we prove that under certain conditions the true support is perfectly recovered within the first iteration.

Amir Ivry, Baruch Berdugo, Israel Cohen

We address voice activity detection in acoustic environments of transients and stationary noises, which often occur in real life scenarios. We exploit unique spatial patterns of speech and non-speech audio frames by independently learning their underlying geometric structure. This process is done through a deep encoder-decoder based neural network architecture. This structure involves an encoder that maps spectral features with temporal information to their low-dimensional representations, which are generated by applying the diffusion maps method. The encoder feeds a decoder that maps the embedded data back into the high-dimensional space. A deep neural network, which is trained to separate speech from non-speech frames, is obtained by concatenating the decoder to the encoder, resembling the known Diffusion nets architecture. Experimental results show enhanced performance compared to competing voice activity detection methods. The improvement is achieved in both accuracy, robustness and generalization ability. Our model performs in a real-time manner and can be integrated into audio-based communication systems. We also present a batch algorithm which obtains an even higher accuracy for off-line applications.

Amir Ivry, Israel Cohen, Baruch Berdugo

We propose a nonlinear acoustic echo cancellation system, which aims to model the echo path from the far-end signal to the near-end microphone in two parts. Inspired by the physical behavior of modern hands-free devices, we first introduce a novel neural network architecture that is specifically designed to model the nonlinear distortions these devices induce between receiving and playing the far-end signal. To account for variations between devices, we construct this network with trainable memory length and nonlinear activation functions that are not parameterized in advance, but are rather optimized during the training stage using the training data. Second, the network is succeeded by a standard adaptive linear filter that constantly tracks the echo path between the loudspeaker output and the microphone. During training, the network and filter are jointly optimized to learn the network parameters. This system requires 17 thousand parameters that consume 500 Million floating-point operations per second and 40 Kilo-bytes of memory. It also satisfies hands-free communication timing requirements on a standard neural processor, which renders it adequate for embedding on hands-free communication devices. Using 280 hours of real and synthetic data, experiments show advantageous performance compared to competing methods.

Amir Ivry, Israel Cohen, Baruch Berdugo

In this paper, we propose a residual echo suppression method using a UNet neural network that directly maps the outputs of a linear acoustic echo canceler to the desired signal in the spectral domain. This system embeds a design parameter that allows a tunable tradeoff between the desired-signal distortion and residual echo suppression in double-talk scenarios. The system employs 136 thousand parameters, and requires 1.6 Giga floating-point operations per second and 10 Mega-bytes of memory. The implementation satisfies both the timing requirements of the AEC challenge and the computational and memory limitations of on-device applications. Experiments are conducted with 161~h of data from the AEC challenge database and from real independent recordings. We demonstrate the performance of the proposed system in real-life conditions and compare it with two competing methods regarding echo suppression and desired-signal distortion, generalization to various environments, and robustness to high echo levels.

Amir Ivry, Israel Cohen, Baruch Berdugo

State-of-the-art deep-learning-based voice activity detectors (VADs) are often trained with anechoic data. However, real acoustic environments are generally reverberant, which causes the performance to significantly deteriorate. To mitigate this mismatch between training data and real data, we simulate an augmented training set that contains nearly five million utterances. This extension comprises of anechoic utterances and their reverberant modifications, generated by convolutions of the anechoic utterances with a variety of room impulse responses (RIRs). We consider five different models to generate RIRs, and five different VADs that are trained with the augmented training set. We test all trained systems in three different real reverberant environments. Experimental results show $20\%$ increase on average in accuracy, precision and recall for all detectors and response models, compared to anechoic training. Furthermore, one of the RIR models consistently yields better performance than the other models, for all the tested VADs. Additionally, one of the VADs consistently outperformed the other VADs in all experiments.

Gal Mishne, Ronen Talmon, Israel Cohen et al.

We consider the analysis of high dimensional data given in the form of a matrix with columns consisting of observations and rows consisting of features. Often the data is such that the observations do not reside on a regular grid, and the given order of the features is arbitrary and does not convey a notion of locality. Therefore, traditional transforms and metrics cannot be used for data organization and analysis. In this paper, our goal is to organize the data by defining an appropriate representation and metric such that they respect the smoothness and structure underlying the data. We also aim to generalize the joint clustering of observations and features in the case the data does not fall into clear disjoint groups. For this purpose, we propose multiscale data-driven transforms and metrics based on trees. Their construction is implemented in an iterative refinement procedure that exploits the co-dependencies between features and observations. Beyond the organization of a single dataset, our approach enables us to transfer the organization learned from one dataset to another and to integrate several datasets together. We present an application to breast cancer gene expression analysis: learning metrics on the genes to cluster the tumor samples into cancer sub-types and validating the joint organization of both the genes and the samples. We demonstrate that using our approach to combine information from multiple gene expression cohorts, acquired by different profiling technologies, improves the clustering of tumor samples.

David Dov, Ronen Talmon, Israel Cohen

In this paper, we address the problem of multiple view data fusion in the presence of noise and interferences. Recent studies have approached this problem using kernel methods, by relying particularly on a product of kernels constructed separately for each view. From a graph theory point of view, we analyze this fusion approach in a discrete setting. More specifically, based on a statistical model for the connectivity between data points, we propose an algorithm for the selection of the kernel bandwidth, a parameter, which, as we show, has important implications on the robustness of this fusion approach to interferences. Then, we consider the fusion of audio-visual speech signals measured by a single microphone and by a video camera pointed to the face of the speaker. Specifically, we address the task of voice activity detection, i.e., the detection of speech and non-speech segments, in the presence of structured interferences such as keyboard taps and office noise. We propose an algorithm for voice activity detection based on the audio-visual signal. Simulation results show that the proposed algorithm outperforms competing fusion and voice activity detection approaches. In addition, we demonstrate that a proper selection of the kernel bandwidth indeed leads to improved performance.

Gal Mishne, Uri Shaham, Alexander Cloninger et al.

Non-linear manifold learning enables high-dimensional data analysis, but requires out-of-sample-extension methods to process new data points. In this paper, we propose a manifold learning algorithm based on deep learning to create an encoder, which maps a high-dimensional dataset and its low-dimensional embedding, and a decoder, which takes the embedded data back to the high-dimensional space. Stacking the encoder and decoder together constructs an autoencoder, which we term a diffusion net, that performs out-of-sample-extension as well as outlier detection. We introduce new neural net constraints for the encoder, which preserves the local geometry of the points, and we prove rates of convergence for the encoder. Also, our approach is efficient in both computational complexity and memory requirements, as opposed to previous methods that require storage of all training points in both the high-dimensional and the low-dimensional spaces to calculate the out-of-sample-extension and the pre-image.

Idan Ram, Michael Elad, Israel Cohen

We propose an image processing scheme based on reordering of its patches. For a given corrupted image, we extract all patches with overlaps, refer to these as coordinates in high-dimensional space, and order them such that they are chained in the "shortest possible path", essentially solving the traveling salesman problem. The obtained ordering applied to the corrupted image, implies a permutation of the image pixels to what should be a regular signal. This enables us to obtain good recovery of the clean image by applying relatively simple 1D smoothing operations (such as filtering or interpolation) to the reordered set of pixels. We explore the use of the proposed approach to image denoising and inpainting, and show promising results in both cases.