Continual learning of neural networks for quality prediction in production using memory aware synapses and weight transfer

Journal of Intelligent Manufacturing - Deep learning-based predictive quality enables manufacturing companies to make data-driven predictions of the quality of a produced product based on process...     

AlphaPilot: autonomous drone racing

Autonomous Robots - This paper presents a novel system for autonomous, vision-based drone racing combining learned data abstraction, nonlinear filtering, and time-optimal trajectory planning. The...     

Quantum computing implementations with neutral particles

We review quantum information processing with cold neutral particles, that is, atoms or polar molecules. First, we analyze the best suited degrees of freedom of these particles for storing quantum information, and then we discuss both single- and two-qubit gate implementations. We focus our discussion mainly on collisional quantum gates, which are best suited for atom-chip-like devices, as well as on gate proposals conceived for optical lattices. Additionally, we analyze schemes both for cold atoms confined in optical cavities and hybrid approaches to entanglement generation, and we show how optimal control theory might be a powerful tool to enhance the speed up of the gate operations as well as to achieve high fidelities required for fault tolerant quantum computation.     

Monte Carlo techniques for phrase-based translation

Recent advances in statistical machine translation have used approximate beam search for NP-complete inference within probabilistic translation models. We present an alternative approach of sampling from the posterior distribution defined by a translation model. We define a novel Gibbs sampler for sampling translations given a source sentence and show that it effectively explores this posterior distribution. In doing so we overcome the limitations of heuristic beam search and obtain theoretically sound solutions to inference problems such as finding the maximum probability translation and minimum risk training and decoding.     

Nano-texturing of magnetic recording sliders via laser ablation

To increase the storage density of hard disk drives, the flying height of the slider needs to be reduced to <10 nm. This requires super-smooth surfaces of the disk and slider. As the roughness decreases, stiction and adhesion are found to increase substantially leading to failures of the head/disk interface. Texturing the slider surface is a well-known approach to this issue. In this study we investigated laser ablation as a potential process for texturing magnetic recording sliders. It was found that straight laser machining caused unwanted re-deposition of material. These deposits could be significantly reduced by using a chemical etching enhanced laser process.     

Continuous-time quantum Monte Carlo impurity solvers

Continuous-time quantum Monte Carlo impurity solvers are algorithms that sample the partition function of an impurity model using diagrammatic Monte Carlo techniques. The present paper describes codes that implement the interaction expansion algorithm originally developed by Rubtsov, Savkin, and Lichtenstein, as well as the hybridization expansion method developed by Werner, Millis, Troyer, et al. These impurity solvers are part of the ALPS-DMFT application package and are accompanied by an implementation of dynamical mean-field self-consistency equations for (single orbital single site) dynamical mean-field problems with arbitrary densities of states.

Program summary

Program title: dmftCatalogue identifier: AEIL_v1_0Program summary URL:http://cpc.cs.qub.ac.uk/summaries/AEIL_v1_0.htmlProgram obtainable from: CPC Program Library, Queen's University, Belfast, N. IrelandLicensing provisions: ALPS LIBRARY LICENSE version 1.1No. of lines in distributed program, including test data, etc.: 899 806No. of bytes in distributed program, including test data, etc.: 32 153 916Distribution format: tar.gzProgramming language: C++Operating system: The ALPS libraries have been tested on the following platforms and compilers:

• • 

  Linux with GNU Compiler Collection (g++ version 3.1 and higher), and Intel C++ Compiler (icc version 7.0 and higher)

• • 

  MacOS X with GNU Compiler (g++ Apple-version 3.1, 3.3 and 4.0)

• • 

  IBM AIX with Visual Age C++ (xlC version 6.0) and GNU (g++ version 3.1 and higher) compilers

• • 

  Compaq Tru64 UNIX with Compq C++ Compiler (cxx)

• • 

  SGI IRIX with MIPSpro C++ Compiler (CC)

• • 

  HP-UX with HP C++ Compiler (aCC)

• • 

  Windows with Cygwin or coLinux platforms and GNU Compiler Collection (g++ version 3.1 and higher)

RAM: 10 MB–1 GBClassification: 7.3External routines: ALPS [1], BLAS/LAPACK, HDF5Nature of problem: (See [2].) Quantum impurity models describe an atom or molecule embedded in a host material with which it can exchange electrons. They are basic to nanoscience as representations of quantum dots and molecular conductors and play an increasingly important role in the theory of “correlated electron” materials as auxiliary problems whose solution gives the “dynamical mean field” approximation to the self-energy and local correlation functions.Solution method: Quantum impurity models require a method of solution which provides access to both high and low energy scales and is effective for wide classes of physically realistic models. The continuous-time quantum Monte Carlo algorithms for which we present implementations here meet this challenge. Continuous-time quantum impurity methods are based on partition function expansions of quantum impurity models that are stochastically sampled to all orders using diagrammatic quantum Monte Carlo techniques. For a review of quantum impurity models and their applications and of continuous-time quantum Monte Carlo methods for impurity models we refer the reader to [2].Additional comments: Use of dmft requires citation of this paper. Use of any ALPS program requires citation of the ALPS [1] paper.Running time: 60 s–8 h per iteration.References:

• [1] 

  A. Albuquerque, F. Alet, P. Corboz, et al., J. Magn. Magn. Mater. 310 (2007) 1187.

• [2] 

  http://arxiv.org/abs/1012.4474, Rev. Mod. Phys., in press.

     

Two‐Level Grids for Ray Tracing on GPUs

We investigate the use of two‐level nested grids as acceleration structure for ray tracing of dynamic scenes. We propose a massively parallel, sort‐based construction algorithm and show that the two‐level grid is one of the structures that is fastest to construct on modern graphics processors. The structure handles non‐uniform primitive distributions more robustly than the uniform grid and its traversal performance is comparable to those of other high quality acceleration structures used for dynamic scenes. We propose a cost model to determine the grid resolution and improve SIMD utilization during ray‐triangle intersection by employing a hybrid packetization strategy. The build times and ray traversal acceleration provide overall rendering performance superior to previous approaches for real time rendering of animated scenes on GPUs.     

Fully-adaptive algorithms for long-lived renaming

Long-lived renaming allows processes to repeatedly get distinct names from a small name space and release these names. This paper presents two long-lived renaming algorithms in which the name a process gets is bounded above by the number of processes currently occupying a name or performing one of these operations. The first algorithm is asynchronous, uses LL/SC objects, and has step complexity that is linear in the number of processes, c, currently getting or releasing a name. The second is synchronous, uses registers and counters, and has step complexity that is polylogarithmic in c. Both tolerate any number of process crashes.     

Precision,Local Search and Unimodal Functions

We investigate the effects of precision on the efficiency of various local search algorithms on 1-D unimodal functions. We present a (1+1)-EA with adaptive step size which finds the optimum in O(log n) steps, where n is the number of points used. We then consider binary (base-2) and reflected Gray code representations with single bit mutations. The standard binary method does not guarantee locating the optimum, whereas using the reflected Gray code does so in Θ((log n)²) steps. A(1+1)-EA with a fixed mutation probability distribution is then presented which also runs in O((log n)²). Moreover, a recent result shows that this is optimal (up to some constant scaling factor), in that there exist unimodal functions for which a lower bound of Ω((log n)²) holds regardless of the choice of mutation distribution. For continuous multimodal functions, the algorithm also locates the global optimum in O((log n)²). Finally, we show that it is not possible for a black box algorithm to efficiently optimise unimodal functions for two or more dimensions (in terms of the precision used).