Search

With the growth of model sizes and the scale of their deployment, their sheer size burdens the infrastructure requiring more network and more storage to accommodate these. While there is a vast model compression literature deleting parts of the model weights for faster inference, we investigate a more traditional type of compression - one that represents the model in a compact form and is coupled with a decompression algorithm that returns it to its original form and size - namely lossless compression. We present ZipNN a lossless compression tailored to neural networks. Somewhat surprisingly, we show that specific lossless compression can gain significant network and storage reduction on popular models, often saving 33% and at times reducing over 50% of the model size. We investigate the source of model compressibility and introduce specialized compression variants tailored for models that further increase the effectiveness of compression. On popular models (e.g. Llama 3) ZipNN shows space savings that are over 17% better than vanilla compression while also improving compression and decompression speeds by 62%. We estimate that these methods could save over an ExaByte per month of network traffic downloaded from a large model hub like Hugging Face., Comment: arXiv admin note: substantial text overlap with arXiv:2404.15198

With the growth of model sizes and scale of their deployment, their sheer size burdens the infrastructure requiring more network and more storage to accommodate these. While there is a vast literature about reducing model sizes, we investigate a more traditional type of compression -- one that compresses the model to a smaller form and is coupled with a decompression algorithm that returns it to its original size -- namely lossless compression. Somewhat surprisingly, we show that such lossless compression can gain significant network and storage reduction on popular models, at times reducing over $50\%$ of the model size. We investigate the source of model compressibility, introduce compression variants tailored for models and categorize models to compressibility groups. We also introduce a tunable lossy compression technique that can further reduce size even on the less compressible models with little to no effect on the model accuracy. We estimate that these methods could save over an ExaByte per month of network traffic downloaded from a large model hub like HuggingFace.

Key-value stores typically leave access control to the systems for which they act as storage engines. Unfortunately, attackers may circumvent such read access controls via timing attacks on the key-value store, which use differences in query response times to glean information about stored data. To date, key-value store timing attacks have aimed to disclose stored values and have exploited external mechanisms that can be disabled for protection. In this paper, we point out that key disclosure is also a security threat -- and demonstrate key disclosure timing attacks that exploit mechanisms of the key-value store itself. We target LSM-tree based key-value stores utilizing range filters, which have been recently proposed to optimize LSM-tree range queries. We analyze the impact of the range filters SuRF and prefix Bloom filter on LSM-trees through a security lens, and show that they enable a key disclosure timing attack, which we call prefix siphoning. Prefix siphoning successfully leverages benign queries for non-present keys to identify prefixes of actual keys -- and in some cases, full keys -- in scenarios where brute force searching for keys (via exhaustive enumeration or random guesses) is infeasible., Comment: Full version of USENIX ATC'23 paper

We study the fundamental problem of selecting optimal features for model construction. This problem is computationally challenging on large datasets, even with the use of greedy algorithm variants. To address this challenge, we extend the adaptive query model, recently proposed for the greedy forward selection for submodular functions, to the faster paradigm of Orthogonal Matching Pursuit for non-submodular functions. The proposed algorithm achieves exponentially fast parallel run time in the adaptive query model, scaling much better than prior work. Furthermore, our extension allows the use of downward-closed constraints, which can be used to encode certain fairness criteria into the feature selection process. We prove strong approximation guarantees for the algorithm based on standard assumptions. These guarantees are applicable to many parametric models, including Generalized Linear Models. Finally, we demonstrate empirically that the proposed algorithm competes favorably with state-of-the-art techniques for feature selection, on real-world and synthetic datasets.

Dimensionality reduction algorithms are standard tools in a researcher's toolbox. Dimensionality reduction algorithms are frequently used to augment downstream tasks such as machine learning, data science, and also are exploratory methods for understanding complex phenomena. For instance, dimensionality reduction is commonly used in Biology as well as Neuroscience to understand data collected from biological subjects. However, dimensionality reduction techniques are limited by the von-Neumann architectures that they execute on. Specifically, data intensive algorithms such as dimensionality reduction techniques often require fast, high capacity, persistent memory which historically hardware has been unable to provide at the same time. In this paper, we present a re-implementation of an existing dimensionality reduction technique called Geometric Multi-Scale Resolution Analysis (GMRA) which has been accelerated via novel persistent memory technology called Memory Centric Active Storage (MCAS). Our implementation uses a specialized version of MCAS called PyMM that provides native support for Python datatypes including NumPy arrays and PyTorch tensors. We compare our PyMM implementation against a DRAM implementation, and show that when data fits in DRAM, PyMM offers competitive runtimes. When data does not fit in DRAM, our PyMM implementation is still able to process the data., Comment: 2021 IEEE High Performance Extreme Computing Conference (HPEC)

Bayesian inference allows machine learning models to express uncertainty. Current machine learning models use only a single learnable parameter combination when making predictions, and as a result are highly overconfident when their predictions are wrong. To use more learnable parameter combinations efficiently, these samples must be drawn from the posterior distribution. Unfortunately computing the posterior directly is infeasible, so often researchers approximate it with a well known distribution such as a Gaussian. In this paper, we show that through the use of high-capacity persistent storage, models whose posterior distribution was too big to approximate are now feasible, leading to improved predictions in downstream tasks.

MCAS (Memory Centric Active Storage) is a persistent memory tier for high-performance durable data storage. It is designed from the ground-up to provide a key-value capability with low-latency guarantees and data durability through memory persistence and replication. To reduce data movement and make further gains in performance, we provide support for user-defined "push-down" operations (known as Active Data Objects) that can execute directly and safely on the value-memory associated with one or more keys. The ADO mechanism allows complex pointer-based dynamic data structures (e.g., trees) to be stored and operated on in persistent memory. To this end, we examine a real-world use case for MCAS-ADO in the handling of enterprise storage system metadata for Continuous Data Protection (CDP). This requires continuously updating complex metadata that must be kept consistent and durable. In this paper, we i.) present the MCAS-ADO system architecture, ii.) show how the CDP use case is implemented, and finally iii.) give an evaluation of system performance in the context of this use case.

The advent of CPU-attached persistent memory technology, such as Intel's Optane Persistent Memory Modules (PMM), has brought with it new opportunities for storage. In 2018, IBM Research Almaden began investigating and developing a new enterprise-grade storage solution directly aimed at this emerging technology. MCAS (Memory Centric Active Storage) defines an evolved network-attached key-value store that offers both near-data compute and the ability to layer enterprise-grade data management services on shared persistent memory. As a converged memory-storage tier, MCAS moves towards eliminating the traditional separation of compute and storage, and thereby unifying the data space. This paper provides an in-depth review of the MCAS architecture and implementation, as well as general performance results., Comment: Revision 1.2

To rectify the problems of electron clouds observed in RHIC and unacceptable ohmic heating for superconducting magnets that can limit future machine upgrades, we started developing a robotic plasma deposition technique for $in-situ$ coating of the RHIC 316LN stainless steel cold bore tubes based on staged magnetrons mounted on a mobile mole for deposition of Cu followed by amorphous carbon (a-C) coating. The Cu coating reduces wall resistivity, while a-C has low SEY that suppresses electron cloud formation. Recent RF resistivity computations indicate that 10 {\mu}m of Cu coating thickness is needed. But, Cu coatings thicker than 2 {\mu}m can have grain structures that might have lower SEY like gold black. A 15-cm Cu cathode magnetron was designed and fabricated, after which, 30 cm long samples of RHIC cold bore tubes were coated with various OFHC copper thicknesses; room temperature RF resistivity measured. Rectangular stainless steel and SS discs were Cu coated. SEY of rectangular samples were measured at room; and, SEY of a disc sample was measured at cryogenic temperatures., Comment: 8 pages, contribution to the Joint INFN-CERN-EuCARD-AccNet Workshop on Electron-Cloud Effects: ECLOUD'12; 5-9 Jun 2012, La Biodola, Isola d'Elba, Italy

We give a new successor data structure which improves upon the index size of the P\v{a}tra\c{s}cu-Thorup data structures, reducing the index size from $O(n w^{4/5})$ bits to $O(n \log w)$ bits, with optimal probe complexity. Alternatively, our new data structure can be viewed as matching the space complexity of the (probe-suboptimal) $z$-fast trie of Belazzougui et al. Thus, we get the best of both approaches with respect to both probe count and index size. The penalty we pay is an extra $O(\log w)$ inter-register operations. Our data structure can also be used to solve the weak prefix search problem, the index size of $O(n \log w)$ bits is known to be optimal for any such data structure. The technical contributions include highly efficient single word indices, with out-degree $w/\log w$ (compared to the $w^{1/5}$ out-degree of fusion tree based indices). To construct such high efficiency single word indices we device highly efficient bit selectors which, we believe, are of independent interest., Comment: 28 pages, full version, extended abstract submitted to MFCS 2013

The plasma lens concept is examined as an alternative to focusing horns and solenoids for a neutrino beam facility. The concept is based on a combined high-current lens/target configuration. Current is fed at an electrode located downstream from the beginning of the target where pion capturing is needed. The current is carried by plasma outside the target. A second plasma lens section, with an additional current feed, follows the target. The plasma is immersed in a relatively small solenoidal magnetic field to facilitate its current profile shaping to optimize pion capture. Simulations of the not yet fully optimized configuration yielded a 25% higher neutrino flux at a detector situated at 3 km from the target than the horn system for the entire energy spectrum and a factor of 2.5 higher flux for neutrinos with energy larger than 3 GeV. A major advantage of plasma lenses is in background reduction. In antineutrino operation, neutrino background is reduced by a factor of close to 3 for the whole spectrum, and for energy larger than 3 GeV, neutrino background is reduced by a factor of 3.6. Plasma lenses have additional advantages: larger axial currents, high signal purity: minimal neutrino background in antineutrino runs. The lens medium consists of plasma, consequently, particle absorption and scattering is negligible. Withstanding high mechanical and thermal stresses in a plasma is not an issue., Comment: 4 pages, 4 figures submitted to Particle Accelerators Conference 2007 (PAC07)

Muon Colliders have unique technical and physics advantages and disadvantages when compared with both hadron and electron machines. They should thus be regarded as complementary. Parameters are given of 4 TeV and 0.5 TeV high luminosity \mu^+ \mu^- colliders, and of a 0.5 TeV lower luminosity demonstration machine. We discuss the various systems in such muon colliders, starting from the proton accelerator needed to generate the muons and proceeding through muon cooling, acceleration and storage in a collider ring. Detector background, polarization, and nonstandard operating conditions are discussed., Comment: 42 pages TeX file (aipproc style), 25 postscript figures. Submitted to the Proceedings of the Symposium on Physics Potential and Development of $\mu^+ \mu^-$ Colliders, San Francisco, CA Dec 13-15, 1995. Edited by D. Cline and D. Sanders