Polynomial multiplication on embedded vector architectures

High-degree, low-precision polynomial arithmetic is a fundamental computational primitive underlying structured lattice based cryptography. Its algorithmic properties and suitability for implementation on different compute platforms is an active area of research, and this article contributes to this line of work: Firstly, we present memory-efficiency and performance improvements for the Toom-Cook/Karatsuba polynomial multiplication strategy. Secondly, we provide implementations of those improvements on Arm® Cortex®-M4 CPU, as well as the newer Cortex-M55 processor, the first M-profile core implementing the M-profile Vector Extension (MVE), also known as Arm® Helium™ technology. We also implement the Number Theoretic Transform (NTT) on the Cortex-M55 processor. We show that despite being singleissue, in-order and offering only 8 vector registers compared to 32 on A-profile SIMD architectures like Arm® Neon™ technology and the Scalable Vector Extension (SVE), by careful register management and instruction scheduling, we can obtain a 3× to 5× performance improvement over already highly optimized implementations on Cortex-M4, while maintaining a low area and energy profile necessary for use in embedded market. Finally, as a real-world application we integrate our multiplication techniques to post-quantum key-encapsulation mechanism Saber

Personalizing Type-Based Facet Ranking Using BERT Embeddings

In Faceted Search Systems (FSS), users navigate the information space through facets, which are attributes or meta-data that describe the underlying content of the collection. Type-based facets (aka t-facets) help explore the categories associated with the searched objects in structured information space. This work investigates how personalizing t-facet ranking can minimize user effort to reach the intended search target. We propose a lightweight personalisation method based on Vector Space Model (VSM) for ranking the t-facet hierarchy in two steps. The first step scores each individual leaf-node t-facet by computing the similarity between the t-facet BERT embedding and the user profile vector. In this model, the user’s profile is expressed in a category space through vectors that capture the users’ past preferences. In the second step, this score is used to re-order and select the sub-tree to present to the user. The final ranked tree reflects the t-facet relevance both to the query and the user profile. Through the use of embeddings, the proposed method effectively handles unseen facets without adding extra processing to the FSS. The effectiveness of the proposed approach is measured by the user effort required to retrieve the sought item when using the ranked facets. The approach outperformed existing personalization baselines.

Group User Profile Modeling Based on Neural Word Embeddings in Social Networks

How to find a user’s interest from similar users a fundamental research problems in socialized recommender systems. Despite significant advances, there exists diversity loss for the majority of recommender systems. With this paper, for expanding the user’s interest, we overcome this challenge by using representative and diverse similar users from followees. First, we model a personal user profile vector via word2vec and term frequency mechanisms. According to user profiles and their follow relationships, we compute content interaction similarity and follow interaction similarity. Second, by combining two kinds of interaction similarity, we calculate the social similarity and discover a diverse group with coverage and dissimilarity. The users in a diverse group can distinguish each other and cover the whole followees, which can model a group user profile (GUP). Then, by tracking the changes of followee set, we heuristically adjust the number of diverse group users and construct an adaptive GUP. Finally, we conduct experiments on Sina Weibo datasets for recommendation, and the experimental results demonstrate that the proposed GUP outperforms conventional approaches for diverse recommendation.

Plant and Animal sub cellular component localization prediction using multiple combination of various machine learning approaches

Membrane proteins are encoded in the genome and functionally important in the living organisms. Information on subcellular localization of cellular proteins has a significant role in the function of cell organelles. Discovery of drug target and system biology between localization and biological function are highly correlated. Therefore, we are predicting the localization of protein using various machine learning approaches. The prediction system based on the integration of the outcome of five sequence based sub-classifiers. The subcellular localization prediction of the final result is based on protein profile vector, which is a result of the sub-classifiers.

Development of High Speed Scanning System for the Fabrication of 3D Metal Parts Using Electron Beam Selective Melting Technology

In this article a method of high speed scanning in Electron Beam Selective Melting (EBSM) technology is presented. Based on the conventional point-to-point vector scanning method, a scanning method nominated profile vector scanning was developed in cooperation with the existing manufacturing system. A scanning speed up to 200m/s was realized using the improved system. 3D parts from 316L metal powder were fabricated using this scanning system and the quality of the fabricated parts was evaluated.

Grains Ultra-Refinement of Materials Imposed by Pulse Laser

Submicron-sized top layer was synthesized on an austenitic stainless steel sample by adopting laser shock processing (LSP). The substructures were characterized and examined by scanning electron microscopy (SEM) and transmission electron microscopy (TEM), moreover, the mechanism of ultra-refinement laser-induced was analyzed. It showed that a refined top surface with average subgrains size of 0.5(m was synthesized by the model of single laser loading, black paint as absorption coating and the peak pressure induced by LSP approximating twice of the dynamic yield strength of target. It indicated that thermoplastic destabilization had happened in heavily localized regions imposed by LSP. Streak-like subgrains were oriented perpendicular to shock wave (Gaussian profile) vector. In order to accommodate plastic strain, streak-like subgrains experienced necking, breaking up and dynamic rotational recrystallization, as a result, the submicron grains were formed on the surface of sample.

Quadratic LYM-type inequalities for intersecting Sperner families

International audience Let $\mathcal{F}\subseteq 2^{[n]}$ be a intersecting Sperner family (i.e. $A \not\subset B, A \cap B \neq \emptyset$ for all $A,B \in \mathcal{F}$) with profile vector $(f_i)_{i=0 \ldots n}$ (i.e. $f_i=|\mathcal{F} \cap \binom{[n]}{i}|$). We present quadratic inequalities in the $f_i$'s which sharpen the previously known linear $\mathrm{LYM}$-type inequalities.

Kernel-Based Discriminant Techniques for Educational Placement

This article considers the problem of educational placement. Several discriminant techniques are applied to a data set from a survey project of science ability. A profile vector for each student consists of five science-educational indictors. The students are intended to be placed into three reference groups: advanced, regular, and remedial. Various discriminant techniques, including Fisher’s discriminant analysis and kernel-based nonparametric discriminant analysis, are compared. The evaluation work is based on the leaving-one-out misclassification score. Results from the five school data sets and 500 bootstrap samples reveal that the kernel-based nonparametric approach with bandwidth selected by cross validation performs reasonably well. The authors regard kernel-based nonparametric procedures as desirable competitors to Fisher’s discriminant rule for handling problems of educational placement.

Higher-Order Motion Perception in Human Visual Cortex: Evidence from fMRI

fMRI was used to investigate human visual cortex responses to higher-order motion stimuli. Acquisition was on a Siemens 1.5 T scanner (T2*, gradient-recalled EPI, TR 3000 ms, TE 84 ms, flip angle 90°, 2 mm × 2 mm voxels, 256 mm FOV, 10 4-mm slices, 54 acquisitions per run). The measured volume included occipital and posterior parietal cortex. T1 scouts and, in some subjects, high resolution T1 volume images were also acquired. Visual stimuli were gamma-corrected movies (480 × 480 pixels), presented by a PowerMac via an LCD projector, shown through the rear of the scanner onto an adjustable mirror fixed above the subject's eyes. Three types of stimuli were used: (1) first-order motion, (2) second-order motion (both radial sine waves on random-dot backgrounds), (3) structure-from-motion consisting of two rotating circular patches (5 deg diameter) within which dots moved in a constant (centripetal) direction, superimposed on randomly moving dots. Three interleaved comparisons were made: stimulus vs blank, first-order vs second-order, and random motion vs structure-from-motion (27 s each phase, 3 repeats). Analysis was based on a correlation coefficient method, after head-motion correction. Initial correlation was with the stimulus profile vector, then with an average BOLD response vector. Voxels with a correlation >0.5 (p<0.0003) were accepted as significant. In all subjects (seven- teen normals), all stimuli evoked bilateral activity in V1/V2 (BA17/18), and in extrastriate area V5/MT (BA37/19). Bilateral activation was also found in areas V3/V3a (BA19) and BA7. A more pronounced activation of area MST/V5a (BA37/39) was found in response to the structure-from-motion stimulus, compared with random motion.[Supported by: Wellcome Trust, Schilling Foundation.]