Samsung (Israel)

Video frame interpolation is a long-studied problem in the video processing field. Recently, deep learning approaches have been applied to this problem, showing impressive results on low-resolution benchmarks. However, these methods do not scale-up favorably to high resolutions. Specifically, when the motion exceeds a typical number of pixels, their interpolation quality is degraded. Moreover, their run time renders them impractical for real-time applications. In this paper we propose IM-Net: an interpolated motion neural network. We use an economic structured architecture and end-to-end training with multi-scale tailored losses. In particular, we formulate interpolated motion estimation as classification rather than regression. IM-Net outperforms previous methods by more than 1.3dB (PSNR) on a high resolution version of the recently introduced Vimeo triplet dataset. Moreover, the network runs in less than 33msec on a single GPU for HD resolution.

The discrete Radon transform (DRT) was defined by Abervuch as an analog of the continuous Radon transform for discrete data. Both the DRT and its inverse are computable in O(n(2) log n) operations for images of size n × n. In this paper, we demonstrate the applicability of the inverse DRT for the reconstruction of a 2-D object from its continuous projections. The DRT and its inverse are shown to model accurately the continuum as the number of samples increases. Numerical results for the reconstruction from parallel projections are presented. We also show that the inverse DRT can be used for reconstruction from fan-beam projections with equispaced detectors.

In this paper, we extend the Blahut-Arimoto algorithm for maximizing Massey's directed information. The algorithm can be used for estimating the capacity of channels with delayed feedback, where the feedback is a deterministic function of the output. In order to maximize the directed information, we apply the ideas from the regular Blahut-Arimoto algorithm, i.e., the alternating maximization procedure, to our new problem. We provide both upper and lower bound sequences that converge to the optimum global value. Our main insight in this paper is that in order to find the maximum of the directed information over a causal conditioning probability mass function, one can use a backward index time maximization combined with the alternating maximization procedure. We give a detailed description of the algorithm, showing its complexity and the memory needed, and present several numerical examples.

Event cameras are sensors with pixels that respond independently and asynchronously to changes in scene illumination. Event cameras have a number of advantages when compared to conventional cameras: low-latency, high temporal resolution, high dynamic range, low power and sparse data output. However, existing event cameras also suffer from comparatively low spatial resolution and are sensitive to noise. Recently, it has been shown that it is possible to reconstruct an intensity frame stream from an event stream. These reconstructions preserve the high temporal rate of the event stream, but tend to suffer from significant artifacts and low image quality due to the shortcomings of event cameras. In this work we demonstrate that it is possible to combine the best of both worlds, by fusing a color frame stream at low temporal resolution and high spatial resolution with an event stream at high temporal resolution and low spatial resolution to generate a video stream with both high temporal and spatial resolutions while preserving the original color information. We utilize a novel event frame interpolation network (EFI-Net), a multi-phase convolutional neural network which fuses the frame and event streams. EFI-Net is trained using only simulated data and generalizes exceptionally well to real-world experimental data. We show that our method is able to interpolate frames where traditional video interpolation approaches fail, while also outperforming event-only reconstructions. We further contribute a new dataset, containing event camera data synchronized with high speed video. This work opens the door to a new application for event cameras, enabling high fidelity fusion with frame based image streams for generation of high-quality high-speed video. The dataset is available at https://drive.google.com/file/d/1UIGVBqNER_5KguYPAu5y7TVg-JlNhz3-/view?usp=sharing

We provide a rigorous mathematical analysis of two communication strategies: soft decode-and-forward (soft-DF) for relay channels and soft partial interference-cancelation (soft-IC) for interference channels. Both strategies involve soft estimation, which assists the decoding process. We consider LDPC codes, not because of their practical benefits, but because of their analytic tractability, which enables an asymptotic analysis similar to random coding methods of information theory. Unlike some works on the closely-related demodulate-and-forward, we assume non-memoryless, code-structure-aware estimation. With soft-DF, we develop simultaneous density evolution to bound the decoding error probability at the destination. This result applies to erasure relay channels. In one variant of soft-DF, the relay applies Wyner-Ziv coding to enhance its communication with the destination, borrowing from compress-and-forward. To analyze soft-IC, we adapt existing techniques for iterative multiuser detection, and focus on binary-input additive white Gaussian noise interference channels. We prove that optimal point-to-point codes are unsuitable for soft-IC, as well as for all strategies that apply partial decoding to improve upon single-user detection and multiuser detection, including Han-Kobayashi.

We developed a new compact indoor-outdoor detector suitable for an embedded digital camera in a mobile phone. The detector works on a Bayer domain image before applying white balance gains. The key idea is to use a small number of photometrical and colorimetrical features typically calculated in the mobile phone cameras for white balance gains evaluation. These features are collected using an annotated image database that was captured using the camera for a variety of indoor and outdoor scenes by different customers. Using this database, a gentle boosting classifier for indoor-outdoor detection is designed and evaluated. An optimal feature subset and optimal number of rounds are selected as well. On a set of 3,176 images, the proposed detector achieves a 1.7% error rate for indoor and 10.8% for outdoor scenes. A comparative study versus a number of known embedded indoor-outdoor detectors shows advantages of the proposed indoor-outdoor detector for mobile phone cameras.

We consider the problem of dynamic spectrum access (DSA) in cognitive wireless networks, consisting of primary users (PUs) and secondary users (SUs), where only partial observations are available at the SUs due to narrowband sensing and transmissions. The network operates in a time-slotted regime, where the traffic patterns of the PUs are modeled as finite-memory Markov chains, that are unknown to the SUs. Since observations are partial, then both channel sensing and access actions affect the throughput. Focusing on the case in which there is a single SU, our objective is to maximize the SU’s long-term throughput. To that aim, we develop a novel algorithm that learns <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">both</i> access and sensing policies via deep Q-learning, dubbed Double Deep Q-network for Sensing and Access (DDQSA). To the best of our knowledge, this is the first work that jointly optimizes both sensing and access policies for DSA via deep Q-learning. Next, we consider wireless networks with access policy which implements a fixed channel hopping dynamics, for which we analytically determine the optimal SU sensing and access policy and its associated throughput. Then, we demonstrate that indeed, the proposed DDQSA algorithm can achieve near-optimal performance for the considered network. Our results show that the proposed DDQSA algorithm learns a policy that implements both sensing and channel access, which significantly outperforms existing approaches, and can achieve the optimal performance in certain scenarios.

Channel coding over arbitrarily permuted parallel channels was first studied by Willems and coworkers. This paper introduces capacity-achieving polar coding schemes for arbitrarily permuted parallel channels where the component channels are memoryless, binary-input, and output-symmetric.

We consider the repair problem for Reed-Solomon (RS) codes, evaluated on an <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbb{F}_{q}$</tex> -linear subspace <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$U \subseteq \mathbb{F}_{q^{m}}$</tex> of dimension <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$d$</tex> , where <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$q$</tex> is a prime power, <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$m$</tex> is a positive integer, and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbb{F}_{q}$</tex> is the Galois field of size <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$q$</tex> . For <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$q &gt; 2$</tex> , we show the existence of a linear repair scheme for the RS code of length <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$n=q^{d}$</tex> and codimension <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$q^{s}, s &lt; d$</tex> , evaluated on <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$U$</tex> , in which each of the <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$n-1$</tex> surviving nodes transmits only <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$r$</tex> symbols of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbb{F}_{q}$</tex> , provided that <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$ms\geq d(m-r)$</tex> . For the case <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$q=2$</tex> , we prove a similar result, with some restrictions on the evaluation linear subspace <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$U$</tex> . Our proof is based on a probabilistic argument, however the result is not merely an existence result; the success probability is fairly large (at least 1/3) and there is a simple criterion for checking the validity of the randomly chosen linear repair scheme.

We consider the repair problem for Reed–Solomon (RS) codes, evaluated on an <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\mathbb {F}_{q}$ </tex-math></inline-formula> -linear subspace <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$U\subseteq \mathbb {F}_{q^{m}} $ </tex-math></inline-formula> of dimension <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$d$ </tex-math></inline-formula> , where <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$q$ </tex-math></inline-formula> is a prime power, <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$m$ </tex-math></inline-formula> is a positive integer, and <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\mathbb {F}_{q}$ </tex-math></inline-formula> is the Galois field of size <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$q$ </tex-math></inline-formula> . For <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$q&gt;2$ </tex-math></inline-formula> , we show the existence of a linear repair scheme for the RS code of length <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$n=q^{d}$ </tex-math></inline-formula> and codimension <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$q^{s}$ </tex-math></inline-formula> , <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$s &lt; d$ </tex-math></inline-formula> , evaluated on <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$U$ </tex-math></inline-formula> , in which each of the <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$n-1$ </tex-math></inline-formula> surviving nodes transmits only <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$r$ </tex-math></inline-formula> symbols of <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\mathbb {F}_{q}$ </tex-math></inline-formula> , provided that <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$ms\geq d(m-r)$ </tex-math></inline-formula> . For the case <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$q=2$ </tex-math></inline-formula> , we prove a similar result, with some restrictions on the evaluation linear subspace <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$U$ </tex-math></inline-formula> . Our proof is based on a probabilistic argument, however the result is not merely an existence result; the success probability is fairly large (at least <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$1/3$ </tex-math></inline-formula> ) and there is a simple criterion for checking the validity of the randomly chosen linear repair scheme. Our result extend the construction of Dau–Milenkovic to the range <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$r &lt; m-s$ </tex-math></inline-formula> , for a wide range of parameters.

Multispectral (MS) imaging systems have a wide range of applications for computer vision and computational photography tasks, but do not yet enjoy widespread adoption due to their prohibitive costs. Recently, advances in the design and fabrication of photonic metamaterials have enabled the development of MS sensors suitable for integration into consumer grade mobile devices. Augmenting existing RGB cameras and their processing algorithms with richer spectral information has the potential to be utilized in many steps of the image processing pipeline, but diverse real world datasets suitable for conducting such research are not freely available. We introduce Beyond RGB <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sup> , a real-world dataset comprising thousands of multispectral and RGB images in diverse real world and lab conditions that is suitable for the development and evaluation of algorithms utilizing multispectral and RGB data. All the scenes in our dataset include a colorimetric reference and a measurement of the spectrum of the scene illuminant. Additionally, we demonstrate the practical use of our dataset through the introduction of a novel illuminant spectral estimation (ISE) algorithm. Our algorithm surpasses the current state-of-the-art (SoTA) by up to 58.6% on established benchmarks and sets a baseline performance on our own dataset.

The capacity of multiuser networks has been a long-standing problem in information theory. Recently, Avestimehr et al. have proposed a deterministic network model to approximate multiuser wireless networks. This model, known as the ADT network model, takes into account the broadcast nature as well as the multiuser interference inherent in the wireless medium. For the types of connections we consider, we show that the results of Avestimehr et al. under the ADT model can be reinterpreted within the algebraic network coding framework introduced by Koetter and Médard. Using this framework, we propose an efficient distributed linear code construction for the deterministic wireless multicast relay network model. Unlike several previous coding schemes, we do not attempt to find flows in the network. Instead, for a layered network, we maintain an invariant where it is required that at each stage of the code construction, certain sets of codewords are linearly independent.

Multi-agent planning in stochastic environments can be framed formally as a decentralized Markov decision problem. Many real-life distributed problems that arise in manufacturing, multi-robot coordination and information gathering scenarios can be formalized using this framework. However, finding the optimal solution in the general case is hard, limiting the applicability of recently developed algorithms. This paper provides a practical approach for solving decentralized control problems when communication among the decision makers is possible, but costly. We develop the notion of communication-based mechanism that allows us to decompose a decentralized MDP into multiple single-agent problems. In this framework, referred to as decentralized semi-Markov decision process with direct communication (Dec-SMDP-Com), agents operate separately between communications. We show that finding an optimal mechanism is equivalent to solving optimally a Dec-SMDP-Com. We also provide a heuristic search algorithm that converges on the optimal decomposition. Restricting the decomposition to some specific types of local behaviors reduces significantly the complexity of planning. In particular, we present a polynomial-time algorithm for the case in which individual agents perform goal-oriented behaviors between communications. The paper concludes with an additional tractable algorithm that enables the introduction of human knowledge, thereby reducing the overall problem to finding the best time to communicate. Empirical results show that these approaches provide good approximate solutions.

Agents with partial observability need to share information to achieve decentralised coordination. However, in resource-constrained systems, indiscriminate communication can create performance bottlenecks by consuming valuable bandwidth. Therefore, there is a tradeoff between the utility attained by communication and its cost. Here we address this tradeoff by developing a novel strategy to make communication selective based on information redundancy; ensuring communication only occurs when necessary, while maintaining acceptable coordination. We apply this strategy to a state-of-the-art communication protocol to evaluate its resource saving benefit in a distributed network routing problem. Furthermore we design a mechanism to adapt its selectivity level to the prevailing resource constraints to ensure further improvements. Empirical studies show our selective strategy achieves relative savings in bandwidth usage of 50-90% with only a 5-10% relative reduction in coordination effectiveness and the adaptive strategy further improves relative bandwidth usage by up to 10% and also relative coordination effectiveness by up to 12% over the non-adaptive approach.

We consider the problem of in-order packet transmission over a cascade of packet-erasure links with acknowledgment (ACK) signals, interconnected by relays. We treat first the case of transmitting a single packet, in which ACKs are unnecessary, over links with independent identically distributed erasures. For this case, we derive tight upper and lower bounds on the probability of arrive failure within an allowed end-to-end communication delay over a given number of links. When the number of links is commensurate with the allowed delay, we determine the maximal ratio between the two—coined information velocity—for which the arrive-failure probability decays to zero; we further derive bounds on the arrive-failure probability when the ratio is below the information velocity, determine the exponential arrive-failure decay rate, and extend the treatment to links with different erasure probabilities. We then elevate all these results for a stream of packets with independent geometrically distributed interarrival times, and prove that the information velocity and the exponential decay rate remain the same for any stationary ergodic arrival process and for deterministic interarrival times. We demonstrate the significance of the derived fundamental limits—the information velocity and the arrive-failure exponential decay rate—by comparing them to simulation results.

Monocular Depth Estimation (MDE) is a fundamental problem in computer vision with numerous applications. Recently, LIDAR-supervised methods have achieved re-markable per-pixel depth accuracy in outdoor scenes. However, significant errors are typically found in the proximity of depth discontinuities, i.e., depth edges, which often hin-der the performance of depth-dependent applications that are sensitive to such inaccuracies, e.g., novel view synthe-sis and augmented reality. Since direct supervision for the location of depth edges is typically unavailable in sparse LIDAR-based scenes, encouraging the MDE model to produce correct depth edges is not straightforward. To the best of our knowledge this paper is the first attempt to address the depth edges issue for LIDAR-supervised scenes. In this work we propose to learn to detect the location of depth edges from densely-supervised synthetic data, and use it to generate supervision for the depth edges in the MDE training. To quantitatively evaluate our approach, and due to the lack of depth edges GT in LIDAR-based scenes, we manually annotated subsets of the KITTI and the DDAD datasets with depth edges ground truth. We demonstrate significant gains in the accuracy of the depth edges with comparable per-pixel depth accuracy on several challenging datasets. Code and datasets are available at htt ps: //github.com/liortalker/MindTheEdge.

We present a simple syndrome-based fast Chase decoding algorithm for Reed–Solomon (RS) codes. Such an algorithm was initially presented by Wu (IEEE Trans. IT, Jan. 2012), building on properties of the Berlekamp–Massey (BM) algorithm. Wu devised a fast polynomial-update algorithm to construct the error-locator polynomial (ELP) as the solution of a certain linear-feedback shift register (LFSR) synthesis problem. This results in a conceptually complicated algorithm, divided into 8 subtly different cases. Moreover, Wu’s polynomial-update algorithm is not immediately suitable for working with vectors of evaluations. Therefore, complicated modifications were required in order to achieve a true “one-pass” Chase decoding algorithm, that is, a Chase decoding algorithm requiring <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$O(n)$ </tex-math></inline-formula> operations per modified coordinate, where <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$n$ </tex-math></inline-formula> is the RS code length. The main result of the current paper is a conceptually simple syndrome-based fast Chase decoding of RS codes. Instead of developing a theory from scratch, we use the well-established theory of Gröbner bases for modules over <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\mathbb {F}_{q}[X]$ </tex-math></inline-formula> (where <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\mathbb {F}_{q}$ </tex-math></inline-formula> is the finite field of <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$q$ </tex-math></inline-formula> elements, for <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$q$ </tex-math></inline-formula> a prime power). The basic observation is that instead of Wu’s LFSR synthesis problem, it is much simpler to consider “the right” minimization problem over a <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">module</i> . The solution to this minimization problem is a simple polynomial-update algorithm that avoids syndrome updates and works seamlessly with vectors of evaluations. As a result, we obtain a conceptually simple algorithm for one-pass Chase decoding of RS codes. Our algorithm is general enough to work with any algorithm that finds a Gröbner basis for the solution module of the key equation as the initial algorithm (including the Euclidean algorithm), and it is not tied only to the BM algorithm.

A common technique for compressing a neural network is to compute the k-rank ℓ2 approximation Ak of the matrix A∈Rn×d via SVD that corresponds to a fully connected layer (or embedding layer). Here, d is the number of input neurons in the layer, n is the number in the next one, and Ak is stored in O((n+d)k) memory instead of O(nd). Then, a fine-tuning step is used to improve this initial compression. However, end users may not have the required computation resources, time, or budget to run this fine-tuning stage. Furthermore, the original training set may not be available. In this paper, we provide an algorithm for compressing neural networks using a similar initial compression time (to common techniques) but without the fine-tuning step. The main idea is replacing the k-rank ℓ2 approximation with ℓp, for p∈[1,2], which is known to be less sensitive to outliers but much harder to compute. Our main technical result is a practical and provable approximation algorithm to compute it for any p≥1, based on modern techniques in computational geometry. Extensive experimental results on the GLUE benchmark for compressing the networks BERT, DistilBERT, XLNet, and RoBERTa confirm this theoretical advantage.

We present a new fast Chase decoding algorithm for binary BCH codes. The new algorithm reduces the complexity in comparison to a recent fast Chase decoding algorithm for Reed–Solomon (RS) codes by the authors (IEEE Trans. IT, 2022), by requiring only a single Kötter iteration per edge of the decoding tree. In comparison to the fast Chase algorithms presented by Kamiya (IEEE Trans. IT, 2001) and Wu (IEEE Trans. IT, 2012) for binary BCH codes, the polynomials updated throughout the algorithm of the current paper typically have a much lower degree. To achieve the complexity reduction, we build on a new isomorphism between two solution modules in the binary case, and on a degenerate case of the soft-decision (SD) version of the Wu list decoding algorithm. Roughly speaking, we prove that when the maximum list size is 1 in Wu list decoding of binary BCH codes, assigning a multiplicity of 1 to a coordinate has the same effect as flipping this coordinate in a Chase-decoding trial. The solution-module isomorphism also provides a systematic way to benefit from the binary alphabet for reducing the complexity in bounded-distance hard-decision (HD) decoding. Along the way, we briefly develop the Gröbner-bases formulation of the Wu list decoding algorithm for binary BCH codes, which is missing in the literature.

As pixel sizes continue to scale down, backside-illuminated (BSI) technology has been recently adopted as a solution to improve pixel SNR performance. In addition, as the application of image sensors widens from digital still cameras to digital camcorders, high-resolution and high-speed operation are required. This paper presents 1/2.33-inch 14.6Mpixel CMOS image sensor employing a 1.4μm BSI pixel architecture with a floating-diffusion (FD) boosting scheme that enables high SNR and high speed read-out.