We present a method to formulate algorithm discovery as program search, and apply it discover optimization algorithms for deep neural network training. leverage efficient search techniques explore an infinite sparse space. To bridge the large generalization gap between proxy target tasks, we also introduce selection simplification strategies. Our discovers simple effective algorithm, $\textbf{Lion}$ ($\textit{Evo$\textbf{L}$ved S$\textbf{i}$gn M$\textbf{o}$me$\textbf{n}$tum}$). It is more...

Effective program repair techniques, which modify faulty programs to fix them with respect given test suites, can substantially reduce the cost of manual debugging. A common approach is iteratively first generate candidate possible bug fixes and then validate against tests until a that passes all found. While this conceptually simple, due potentially high number candidates need be generated compiled tested, existing techniques embody have relatively low effectiveness, especially for faults...

A high-efficiency inverse design of “digital” subwavelength nanophotonic devices using the adjoint method is proposed. We a single-mode 3 dB power divider and dual-mode demultiplexer to demonstrate efficiency proposed approach, called digitized method, for single- dual-object optimization, respectively. The optimization comprises three stages: 1) continuous variation an “analog” pattern; 2) forced permittivity biasing “quasi-digital” 3) multilevel digital pattern. Compared with conventional...

In Automatic Driving System (ADS) and Driver Assistance (DAS), object detection plays a vital part. Nevertheless, existing real-time models for tiny vehicle objects have the problems of low precision poor performance. To solve these issues, we propose novel model based on You Only Look Once Version 2 (YOLO-v2) deep learning framework objects, called Optimized (O-YOLO-v2). proposed model, new structure is introduced to strengthen feature extraction ability network by adding convolution layers...

In this study, electrically insulating polyolefin elastomer (POE)-based phase change materials (PCMs) comprising alumina (Al2O3) and graphene nanoplatelets (GNPs) are prepared using a conventional injection moulding technique, which exhibits promising applications for solar energy storage due to the reduced interfacial thermal resistance, excellent stability, proficient photo-thermal conversion efficiency. A synergistic interplay between Al2O3 GNPs is observed, facilitates establishment of...

Deep Neural Networks (DNN) are increasingly used in a variety of applications, many them with substantial safety and security concerns. This paper introduces DeepCheck, new approach for validating DNNs based on core ideas from program analysis, specifically symbolic execution. The idea is to translate DNN into an imperative program, thereby enabling analysis assist validation. A basic translation however creates programs that very complex analyze. DeepCheck novel techniques lightweight...

Large-scale Ads recommendation and auction scoring models at Google scale demand immense computational resources. While specialized hardware like TPUs have improved linear algebra computations, bottlenecks persist in large-scale systems. This paper proposes solutions for three critical challenges that must be addressed efficient end-to-end execution a widely used production infrastructure: (1) Input Generation Ingestion Pipeline: Efficiently transforming raw features (e.g., "search query")...

We present two novel approaches for automated testing of models written in Alloy – a well-known declarative, first-order language that is supported by fully automatic SAT-based analysis engine. The first approach introduces test generation and embodied three techniques create suites the traditional spirit black-box, white-box, mutation-based testing. second mutation defines how to mutants models, compute results, check equivalent using SAT. build on theoretical foundation defined previously...

Automated program repair is an active research area. However, existing focuses mostly on imperative code, e.g. in Java. In this paper, we study the problem of repairing declarative models Alloy -- a first order relational logic with transitive closure. We introduce ARepair, technique for models. ARepair follows spirit traditional automated techniques. Specifically, takes as input faulty model and test suite that contains some failing test, outputs repaired correct respect to given tests....

This paper introduces DeepCheck, a new approach for validating Deep Neural Networks (DNNs) based on core ideas from program analysis, specifically symbolic execution. DeepCheck implements techniques lightweight analysis of DNNs and applies them in the context image classification to address two challenging problems: 1) identification important pixels (for attribution adversarial generation); 2) creation attacks. Experimental results using MNIST data-set show that DeepCheck's provides...

Manually locating and removing bugs in faulty program is often tedious error-prone. A common automated repair approach called generate-and-validate (G&V) iteratively creates candidate fixes, compiles them, runs these candidates against the given tests. This can be costly due to a large number of re-compilations re-executions program. To tackle this limitation, recent work introduced SketchFix that tightly integrates generation validation phases, utilizes runtime behaviors substantially prune...

Fault localization is a popular research topic and many techniques have been proposed to locate faults in imperative code, e.g. C Java. In this paper, we focus on the problem of fault for declarative models Alloy - first order relational logic with transitive closure. We introduce AlloyFL <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">hy</sub> , technique faulty which leverages multiple test formulas. brings traditional spectrum-based...

Regression testing checks that recent project changes do not break previously working functionality. Although important, regression is costly when are frequent. test selection (RTS) optimizes by running only tests whose results might be affected a change. Traditionally, RTS collects dependencies (e.g., on files) for each and skips the tests, at new revision, did Existing techniques differentiate behavior-preserving transformations (i.e., refactorings) from other code changes. As result, run...

Small intrusions dominated by olivine- and pyroxene-rich cumulates are well known to be favourable hosts magmatic Ni-Cu-(Platinum Group Element - PGE) sulfide mineralization. Such common in a variety of settings around the world, but only very small proportion contain economically exploitable sulfides; these tend conduit or chonolith style. If prospectivity could discriminated from sparse sampling at early exploration stages, then discovery rate for deposits this type improved. To end,...

The rotary energy recovery device (RERD) plays an important role in reverse osmosis (RO) desalination; however, few investigations on the formation and influence of lateral force RERD rotor have been published. transient characteristics its relationship with pressure distribution fluctuation clearance were analyzed via computational fluid dynamics (CFD) simulation. quantified under different working conditions. eccentricity rotor, resistance torque decrease speed due to simulated they found...

Software models help improve the reliability of software systems: can convey requirements, and analyze design implementation properties. A key strength Alloy, a commonly used first-order modeling language, is Alloy Analyzer tool-set. The allows users to execute commands over by leveraging fully automatic SAT-based analysis engine. However, prior introduction AUnit - testing framework for had rely on ad-hoc practices validate their models. In this paper, we present our efforts establish...

Creating models of software systems and analyzing the helps develop more reliable systems. A well-known modeling tool-set is embodied by declarative language Alloy its automatic SAT-based analyzer. Recent work introduced a novel approach to testing validate their correctness in spirit traditional testing: Unit defined foundations (unit tests, test execution, model coverage) for Alloy, MuAlloy mutation (mutation operators, mutant generation, equivalent checking) Alloy. This tool paper...

Deep Neural Networks (DNN) are increasingly used in a variety of applications, many them with serious safety and security concerns. This paper describes DeepCheck, new approach for validating DNNs based on core ideas from program analysis, specifically symbolic execution. DeepCheck implements novel techniques lightweight analysis applies to address two challenging problems DNN analysis: 1) identification important input features 2) leveraging those create adversarial inputs. Experimental...

Scale-invariant feature transform (SIFT) is a popular pattern recognition method in 2D-image because it can abstracts the features which are invariant to rotation, scale zooming, brightness changing. So demonstrates certain stability objects subjected view point changing and noise distribution. However, dimension of SIFT descriptors too high, its runtime long. Aiming at this disadvantage, paper propose new generate descriptor based on hierarchical region treat different regions differently....