Video conferencing tools have seen a significant increase in usage in the past few years but they still consume a significant bandwidth of $\sim 100$ Kbps to a few Mbps. In this work, we present Txt2Vid-Web: a practical, web-based, low bandwidth, video conferencing platform building upon the Txt2Vid work [1]. We introduce multiple improvements over the existing Txt2Vid framework – implementing it on browser application stack making it much more accessible and portable, reducing the implementation complexity of the plat-form via WebGL, and implementing a new WebGL shader for ConvTranspose in ONNX runtime – thereby enabling it to run on web-browsers of modern laptops (Fig. 1a). We use WebRTC to establish peer-to-peer data channels over network connections and utilize SDP’s multimedia negotiation scheme over SRTP connections, allowing our platform to provide high-quality video calls for the majority of connections and fall back to Txt2Vid when bandwidth limitations overconstrain SDP’s chosen codecs. We verified our plat-form via subjective study $(n =126)$ consisting of comparison of five different audio-video (AV) contents compressed via standard codecs and Txt2Vid-Web. We choose bitrates of {6 kbps, 10 kbps} for encoding the audio and bitrates of {15 kbps, 35 kbps, 100 kbps} for encoding the video using standard AV codec. Results show that at similar quality of experience our platform requires $100 - 500 \times$ less bandwidth than H.264 and VP9 as video codec and OPUS as audio codec (Fig. 1b). We envision our platform can open up many novel applications. Towards this end, we also open-source both our new tool as a Github repository (https://github.com/tpulkit/txt2vid_browser), and our subjective study dataset (https://tinyurl.com/SubjectiveStudyDataset).

COVID-19 has made video communication one of the most important modes of information exchange. While extensive research has been conducted on the optimization of the video streaming pipeline, in particular the development of novel video codecs, further improvement in the video quality and latency is required, especially under poor network conditions. This paper proposes an alternative to the conventional codec through the implementation of a keypoint-centric encoder relying on the transmission of keypoint information from within a video feed, as shown in Figure 1. The decoder uses the streamed keypoints to generate a reconstruction preserving the semantic features in the input feed. Focusing on video calling applications, we detect and transmit the body pose and face mesh information through the network, which are displayed at the receiver in the form of animated puppets. Using efficient pose and face mesh detection in conjunction with skeleton-based animation, we demonstrate a prototype requiring lower than 35 kbps bandwidth, an order of magnitude reduction over typical video calling systems. The added computational latency due to the mesh extraction and animation is below 120ms on a standard laptop, showcasing the potential of this framework for real-time applications. The code for this work is available at https://github.com/shubhamchandak94/digital-puppetry/ and the full version is available on arXiv [1].

Bid Shading has become increasingly important in Online Advertising, with a large amount of commercial [4,12,13,29] and research work [11,20,28] recently published. Most approaches for solving the bid shading problem involve estimating the probability of win distribution, and then maximizing surplus [28]. These generally use parametric assumptions for the distribution, and there has been some discussion as to whether Log-Normal, Gamma, Beta, or other distributions are most effective [8,38,41,44]. In this paper, we show evidence that online auctions generally diverge in interesting ways from classic distributions. In particular, real auctions generally exhibit significant structure, due to the way that humans set up campaigns and inventory floor prices [16,26]. Using these insights, we present a nonparametric method for Bid Shading which enables the exploitation of this deep structure. The algorithm has low time and space complexity, and is designed to operate within the challenging millisecond Service Level Agreements of Real-Time Bid Servers. We deploy it in one of the largest Demand Side Platforms in the United States, and show that it reliably out-performs best in class Parametric benchmarks. We conclude by suggesting some ways that the best aspects of parametric and nonparametric approaches could be combined.

Time series data compression is emerging as an important problem with the growth in IoT devices and sensors. Due to the presence of noise in these datasets, lossy compression can often provide significant compression gains without impacting the performance of downstream applications. In this work, we propose an error-bounded lossy compressor, LFZip, for multivariate floating-point time series data that provides guaranteed reconstruction up to user-specified maximum absolute error. The compressor is based on the prediction-quantization-entropy coder framework and benefits from improved prediction using linear models and neural networks. We evaluate the compressor on several time series datasets where it outperforms the existing state-of-the-art error-bounded lossy compressors. The code and data are available at https://github.com/shubhamchandak94/LFZip.

Lossy image compression has been studied extensively in the context of typical loss functions such as RMSE, MS-SSIM, etc. However, it is not well understood what loss function might be most appropriate for human perception. Furthermore, the availability of massive public image datasets appears to have hardly been exploited in image compression. In this work, we perform compression experiments in which one human describes images to another, using publicly available images and text instructions. These image reconstructions are rated by human scorers on the Amazon Mechanical Turk platform and compared to reconstructions obtained by existing image compressors. In our experiments, the humans outperform the state of the art compressor WebP in the MTurk survey on most images, which shows that there is significant room for improvement in image compression for human perception. The images, results and additional data is available at https://compression.stanford.edu/human-compression.

The affordability of DNA sequencing has led to unprecedented volumes of genomic data. These data must be stored, processed, and analyzed. The most popular format for genomic data is the SAM format, which contains information such as alignment, quality values, etc. These files are large (on the order of terabytes), which necessitates compression. In this work we propose a new reference-based compressor for SAM files, which can accommodate different levels of compression, based on the specific needs of the user. In particular, the proposed compressor GeneComp allows the user to perform lossy compression of the quality scores, which have been proven to occupy more than half of the compressed file (when losslessly compressed). We show that the proposed compressor GeneComp overall achieves better compression ratios than previously proposed algorithms when working on lossless mode.

Summary form only given. We propose a method and algorithm for lossless compression of tabular data - including, for example, machine learning datasets, server logs and genomic datasets. Superior compression ratios are achieved by exploiting dependencies between the fields (or "features") in the dataset. The algorithm compresses the records w.r.t. a probabilistic graphical model - specifically an optimized forest, where each feature is a node. The work extends a method known as a Chow-Liu tree by incorporating a more accurate correction term to the cost function, which corresponds to the size required to describe the model itself. Additional features of the algorithm are efficient coding of the metadata (such as probability distributions), as well as data relabeling in order to cope with large datasets and alphabets. We test the algorithm on several datasets, and demonstrate an improvement in the compression rates of between 2X and 5X compared to gzip. The larger improvements are observed for very large datasets, such as the Criteo click prediction dataset which was published as part of a recent Kaggle competition.

Massive amounts of sequencing data are being generated thanks to advances in sequencing technology and a dramatic drop in the sequencing cost. Storing and sharing this large data has become a major bottleneck in the discovery and analysis of genetic variants that are used for medical inference. As such, lossless compression of this data has been proposed. Of the compressed data, more than 70% correspond to quality scores, which indicate the sequencing machine reliability when calling a particular basepair. Thus, to further improve the compression performance, lossy compression of quality scores is emerging as the natural candidate. Since the data is used for genetic variants discovery, lossy compressors for quality scores are analyzed in terms of their rate-distortion performance, as well as their effect on the variant callers. Previously proposed algorithms do not do well under all performance metrics, and are hence unsuitable for certain applications. In this work we propose a new lossy compressor that first performs a clustering step, by assuming all the quality scores sequences come from a mixture of Markov models. Then, it performs quantization of the quality scores based on the Markov models. Each quantizer targets a specific distortion to optimize for the overall rate-distortion performance. Finally, the quantized values are compressed by an entropy encoder. We demonstrate that the proposed lossy compressor outperforms the previously proposed methods under all analyzed distortion metrics. This suggests that the effect that the proposed algorithm will have on any downstream application will likely be less noticeable than that of previously proposed lossy compressors. Moreover, we analyze how the proposed lossy compressor affects Single Nucleotide Polymorphism (SNP) calling, and show that the variability introduced on the calls is considerably smaller than the variability that exists between different methodologies for SNP calling.

Massive amounts of sequencing data are being generated thanks to advances in sequencing technology and a dramatic drop in the sequencing cost. Much of the raw data are comprised of nucleotides and the corresponding quality scores that indicate their reliability. The latter are more difficult to compress and are themselves noisy. Lossless and lossy compression of the quality scores has recently been proposed to alleviate the storage costs, but reducing the noise in the quality scores has remained largely unexplored. This raw data is processed in order to identify variants; these genetic variants are used in important applications, such as medical decision making. Thus improving the performance of the variant calling by reducing the noise contained in the quality scores is important. We propose a denoising scheme that reduces the noise of the quality scores and we demonstrate improved inference with this denoised data. Specifically, we show that replacing the quality scores with those generated by the proposed denoiser results in more accurate variant calling in general. Moreover, a consequence of the denoising is that the entropy of the produced quality scores is smaller, and thus significant compression can be achieved with respect to lossless compression of the original quality scores. We expect our results to provide a baseline for future research in denoising of quality scores. The code used in this work as well as a Supplement with all the results are available at http://web.stanford.edu/~iochoa/DCCdenoiser_CodeAndSupplement.zip.

We consider the problem of compressing discrete memory less data sequences for the purpose of similarity identification, first studied by Ahlswede et al. (1997). In this setting, a source sequence is compressed, where the goal is to be able to identify whether the original source sequence is similar to another given sequence (called the query sequence). There is no requirement that the source will be reproducible from the compressed version. In the case where no false negatives are allowed, a compression scheme is said to be reliable if the probability of error (false positive) vanishes as the sequence length grows. The minimal compression rate in this sense, which is the parallel of the classical rate distortion function, is called the identification rate. The rate at which the error probability vanishes is measured by its exponent, called the identification exponent (which is the analog of the classical excess distortion exponent). While an information-theoretic expression for the identification exponent was found in past work, it is uncomputable due to a dependency on an auxiliary random variable with unbounded cardinality. The main result of this paper is a cardinality bound on the auxiliary random variable in the identification exponent, thereby making the quantity computable (solving the problem that was left open by Ahlswede et al.). The new proof technique relies on the fact that the Lagrangian in the optimization problem (in the expression for the exponent) can be decomposed by coordinate (of the auxiliary random variable). Then a standard Caratheodory - style argument completes the proof.

We consider compression of sequences in a database so that similarity queries can be performed efficiently in the compressed domain. The fundamental limits for this problem setting, which characterize the trade off between compression rate and reliability of the answers to the queries, have been characterized in past work. However, how to approach these limits in practice has remained largely unexplored. Recently, we proposed a scheme for this task that is based on existing lossy compression algorithms, for the general case where the similarity measure satisfies a triangle inequality. Although it was shown that it achieves the fundamental limits for some cases, it is suboptimal in general. In this paper we propose a new scheme that also uses lossy compression algorithms as a building block, but with a carefully chosen distortion measure that is different than the one defining the similarity between sequences. The new scheme significantly improves the compression rate compared to the previously proposed scheme in many cases. For example, for binary sources and Hamming similarity measure, simulation results show a compression rate close to the fundamental limit, and an improvement over the previously proposed scheme of up to 55% (for the same reliability). The results shed light on the fact that compression for similarity identification is inherently different than classical lossy compression.

The problem of performing similarity queries on compressed data is considered. We study the fundamental tradeoff between compression rate, sequence length, and reliability of queries performed on compressed data. For a Gaussian source and quadratic similarity criterion, we show that queries can be answered reliably if and only if the compression rate exceeds a given threshold - the identification rate - which we explicitly characterize. When compression is performed at a rate greater than the identification rate, responses to queries on the compressed data can be made exponentially reliable. We give a complete characterization of this exponent, which is analogous to the error and excess-distortion exponents in channel and source coding, respectively. For a general source, we prove that the identification rate is at most that of a Gaussian source with the same variance. Therefore, as with classical compression, the Gaussian source requires the largest compression rate. Moreover, a scheme is described that attains this maximal rate for any source distribution.

Motivated by the Markov chain Monte Carlo (MCMC) relaxation method of Jalali and Weissman, we propose a lossy compression algorithm for continuous amplitude sources that relies on a finite reproduction alphabet that grows with the input length. Our algorithm asymptotically achieves the optimum rate distortion (RD) function universally for stationary ergodic continuous amplitude sources. However, the large alphabet slows down the convergence to the RD function, and is thus an impediment in practice. We thus propose an MCMC-based algorithm that uses a (smaller) adaptive reproduction alphabet. In addition to computational advantages, the reduced alphabet accelerates convergence to the RD function, and is thus more suitable in practice.

We present a new lossy compressor for discrete sources. For coding a source sequence xn, the encoder starts by assigning a certain cost to each reconstruction sequence. It then finds the reconstruction that minimizes this cost and describes it losslessly to the decoder via a universal lossless compressor. The cost of a sequence is given by a linear combination of its empirical probabilities of some order k+1 and its distortion relative to the source sequence. The linear structure of the cost in the empirical count matrix allows the encoder to employ a Viterbi-like algorithm for obtaining the minimizing reconstruction sequence simply. We identify a choice of coefficients for the linear combination in the cost function which ensures that the algorithm universally achieves the optimum rate-distortion performance of any Markov source in the limit of large n, provided k is increased as o(log n).

Recent work by Orlitsky et al. (2004) has motivated the study of pattern sequences and their compressibility properties. Emphasis in this recent line of work has been on compressing pattern sequences under uncertainty in the source that has generated them, thus focusing on universal schemes and their redundancy. Our interest in this work is in the entropy rate of pattern sequences of stochastic processes, and its relationship to the entropy rate of the original process. We give a complete characterization of this relationship for i.i.d. processes over arbitrary alphabets, stationary and ergodic: processes over discrete alphabets, as well as more general processes that can be represented as the output of an additive white-noise channel. For cases where the entropy rate of the pattern process is infinite, we characterize the possible growth rate of the block entropy.

In the discrete filtering problem, a data sequence over a finite alphabet is assumed to be corrupted by a discrete memoryless channel. The goal is to reconstruct the clean sequence, with as high a fidelity as possible, by way of causal processing of the noisy sequence alone, with the reconstruction at time t depending only on noisy observations occurring no later than t. A universal version of this problem in which no assumptions are made about the distribution of the clean data, which may even be nonstochastic is studied. Using techniques from universal data compression, in particular, the incremental parsing rule of LZ78, and derives a practical and efficient algorithms for the universal filtering of discrete sources. A finite-memory filter of order k has the property that the reconstruction at any time t is a time-invariant function only of noisy observations occurring between times t-k and t, inclusive. The universal filtering algorithms perform essentially as well, in an expected sense (with respect to the noise process), as the best finite-memory filter of any fixed order, determined with full knowledge of the actual clean data sequence, for all such data sequences. Also consider more general finite-state filters and show that any such filter is arbitrarily well approximated by a finite-memory filter of growing order, thereby establishing the universality of the proposed algorithms with respect to this larger class. This result can be viewed as the filtering analogue of the well known optimality of LZ78 relative to the class of finite-state compressors.