SDStark C. Draper

Stark C. Draper

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Stark C. Draper has a total of 67 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    AstroSURE: Learning to Remove Noise from Astronomical Images Without Ground Truth Data

    by Omid Vaheb, Sébastien Fabbro, Stark C. Draper

    In astronomical imaging, the low photon count of exposures necessitates extensive post-processing steps including contamination removal and denoising. This paper introduces AstroSURE, a framework for evaluating and comparing deep-learning denoising methods that can be trained without clean ground-truth images, and assesses their utility for detection-oriented analysis of astronomical data. We adapt and compare Noise2Noise, Stein’s Unbiased Risk Estimator, and blind-spot-based methods using synthetic data and real observations from the Hubble Space Telescope (HST) and the terrestrial Canada–France–Hawaii Telescope (CFHT). Performance is evaluated using object-detection metrics, including correct detection rate and false alarm rate (FAR), together with image-based metrics and pixel-distribution diagnostics. The results show that these methods can improve faint-source detectability relative to the original noisy images, with encouraging gains on HST data after domain-consistent initialization, while transfer to CFHT data is more limited, highlighting the importance of instrument/domain similarity for unsupervised adaptation.

  • Soft Demapping of Spherical Codes From Cartesian Powers of PAM Constellations

    by Reza Rafie Borujeny, Susanna E. Rumsey, Stark C. Draper, Frank R. Kschischang

    For applications in concatenated coding for optical communications systems, we examine soft-demapping of short spherical codes constructed as constant-energy shells of the Cartesian power of pulse amplitude modulation constellations. These are unions of permutation codes having the same average power. We construct a list decoder for permutation codes by adapting Murty’s algorithm, which is then used to determine mutual information curves for these permutation codes. In the process, we discover a straightforward expression for determining the likelihood of large subcodes of permutation codes. We refer to these subcodes, obtained by all possible sign flips of a given permutation codeword, as orbits. We introduce a simple process, which we call orbit demapping with frozen symbols, that allows us to extract soft information from noisy permutation codewords. In a sample communication system with probabilistic amplitude shaping protected by a standard low-density parity-check code that employs short permutation codes, we demonstrate that orbit demapping with frozen symbols provides a gain of about 0.3 dB in signal-to-noise ratio compared to the traditional symbol-by-symbol demapping. By using spherical codes composed of unions of permutation codes, we can increase the input entropy compared to using permutation codes alone. In one scheme, we consider a union of a small number of permutation codes. In this case, orbit demapping with frozen symbols provides about 0.2 dB gain compared to the traditional method. In another scheme, we use all possible permutations to form a spherical code that exhibits a computationally feasible trellis representation. The soft information obtained using the BCJR algorithm outperforms the traditional symbol-by-symbol method by 0.1 dB. Overall, using the spherical codes containing all possible permutation codes of the same average power and the BCJR algorithm, a gain of 0.5 dB is observed compared with the case of using one permutation code with the symbol-by-symbol demapping. Comparison of the achievable information rates of bit-metric decoding verifies the observed gains.

  • Exploiting Stragglers in Distributed Computing Systems With Task Grouping

    by Tharindu Adikari, Haider Al-Lawati, Jason Lam, Zhenhua Hu, Stark C. Draper

    We consider the problem of stragglers in distributed computing systems. Stragglers, which are compute nodes that unpredictably slow down, often increase the completion times of tasks. One common approach to mitigating stragglers is work replication, where only the first completion among replicated tasks is accepted, discarding the others. However, discarding work leads to resource wastage. In this paper, we propose a method for exploiting the work completed by stragglers rather than discarding it. The idea is to increase the granularity of the assigned work, and to increase the frequency of worker updates. We show that the proposed method reduces the completion time of tasks via experiments performed on a simulated cluster as well as on Amazon EC2 with Apache Hadoop


  • Parallelized Block-Based Distribution Matching

    by Maxim Goukhshtein, Stark C. Draper, Jeebak Mitra

    Probabilistic shaping (PS) techniques, when combined with forward error correction, enable reliable transmission at rates close to capacity by inducing a favorable probability distribution on channel input symbols. In this paper, we introduce the parallelized block-based distribution matching (PB-DM) shaping architecture. To facilitate high-throughput PS, the method operates by combining blocks of bits, generated in parallel by multiple binary distribution matchers, and mapping them to blocks of symbols from a non-binary alphabet. The sizes of bit and symbol blocks, and the mapping rule between such blocks, are key design parameters. The parameters bestow the PB-DM architecture with a high degree of customizability, which distinguishes it from other PS schemes. In particular, different PB-DM instances – each characterized by a particular trade-off between shaping performance, latency and memory footprint – can be constructed. Considering a scenario in which constraints are imposed on the allowable latency and memory usage, we propose a heuristic approach to determine design parameters that produce well-performing PB-DM schemes that satisfy the given constraints. We present simulation results over the additive white Gaussian noise channel, demonstrating the performance-complexity trade-offs realizable by different short-blocklength PB-DM designs, and compare them against the trade-offs achieved by other commonly used schemes.

  • Controlled Privacy Leakage Propagation Throughout Overlapping Grouped Learning

    by Shahrzad Kiani, Franziska Boenisch, Stark C. Draper

    Federated Learning (FL) is the standard protocol for collaborative learning. In FL, multiple workers jointly train a shared model. They exchange model updates calculated on their data, while keeping the raw data itself local. Since workers naturally form groups based on common interests and privacy policies, we are motivated to extend standard FL to reflect a setting with multiple, potentially overlapping groups. In this setup where workers can belong and contribute to more than one group at a time, complexities arise in understanding privacy leakage and in adhering to privacy policies. To address the challenges, we propose differential private overlapping grouped learning (DP-OGL), a novel method to implement privacy guarantees within overlapping groups. Under the honest-but-curious threat model, we derive novel privacy guarantees between arbitrary pairs of workers. These privacy guarantees describe and quantify two key effects of privacy leakage in DP-OGL: propagation delay, i.e., the fact that information from one group will leak to other groups only with temporal offset through the common workers and information degradation, i.e., the fact that noise addition over model updates limits information leakage between workers. Our experiments show that applying DP-OGL enhances utility while maintaining strong privacy compared to standard FL setups.


  • Exploiting Parity-Polytope Geometry in Approximate and Randomized Scheduled ADMM-LP Decoding

    by Amirreza Asadzadeh, Anthony Ho, Frank R. Kschischang, Stark C. Draper

    We present two strategies to reduce the complexity of the alternating direction method of multipliers when applied to linear programming (ADMM-LP) decoding of low-density parity-check codes. First, to address the high complexity of computing a projection onto the parity polytope, the complexity bottleneck of ADMM-LP decoding, we propose the sparse affine projection algorithm (SAPA). SAPA projects onto the affine hull of χ ≤ d nearby local codewords where the check degree is d and where χ can be significantly smaller than d . Unlike exact projection, SAPA does not require a water-filling process, and thus can be implemented with lower per-iteration complexity. Second, to reduce the number of effective iterations needed for ADMM-LP decoding, we propose a randomized layered scheduling framework. Rather than updating checks in round-robin fashion in each iteration, more “problematic” checks have a higher probability of being updated. The probability mass function that governs the selection of which checks to update is based upon the location of replica vectors inside (or on) the parity polytope. The resultant decoder converges significantly faster under this randomized scheduling than under round-robin scheduling. This makes it well suited for use in applications that limit the number of iterations.

  • People, Ideas, Impact

    by Stark C. Draper

    It is now spring in Toronto. The organization of our international symposium is proceedings apace. As president, one is involved in/has an observer’s view into many aspects of the organization of the symposium. It is gratifying and impressive to witness the hard work and commitment dedicated to the ISIT by so many society members. A myriad of tasks must be completed and an astronomical number of detailed and thoughtful decisions must be made behind the scenes—arranging the overall symposium, forming the technical program, conceiving and arranging the associated events, on and on—all together these build to the meeting we all enjoy and benefit from so much. Thanks to all the all the organizers and volunteers!

  • Open Access

    Two-terminal source coding with common sum reconstruction

    by Tharindu Adikari, Stark Draper

    We present the problem of two-terminal source coding with Common Sum Reconstruction (CSR). Consider two terminals, each with access to one of two correlated sources. Both terminals want to reconstruct the sum of the two sources under some average distortion constraint, and the reconstructions at two terminals must be identical with high probability. In this paper, we develop inner and outer bounds to the achievable rate distortion region of the CSR problem for a doubly symmetric binary source. We employ existing achievability results for Steinberg's common reconstruction and Wyner-Ziv's source coding with side information problems, and an achievability result for the lossy version of Korner-Marton's modulo-two sum computation problem.

  • Successive Approximation Coding for Distributed Matrix Multiplication

    by Shahrzad Kiani, Stark C. Draper

    Coded distributed computing was recently introduced to mitigate the effect of stragglers on distributed computing systems. This paper combines ideas of approximate and coded computing to further accelerate computation. We propose successive approximation coding (SAC) techniques that realize a tradeoff between accuracy and speed, allowing the distributed computing system to produce approximations that increase in accuracy over time. If a sufficient number of compute nodes finish their tasks, SAC exactly recovers the desired computation. We theoretically provide design guidelines for our SAC techniques, and numerically show that SAC achieves a better accuracy-speed tradeoff in comparison with previous methods.