File descriptor | EPFL Graph Search

In Unix and Unix-like computer operating systems, a file descriptor (FD, less frequently fildes) is a process-unique identifier (handle) for a or other input/output resource, such as a pipe or network socket. File descriptors typically have non-negative integer values, with negative values being reserved to indicate "no value" or error conditions. File descriptors are a part of the POSIX API. Each Unix process (except perhaps daemons) should have three standard POSIX file descriptors, corresponding to the three standard streams: Overview In the traditional implementation of Unix, file descriptors index into a per-process maintained by the kernel, that in turn indexes into a system-wide table of files opened by all processes, called the . This table records the mode with which the file (or other resource) has been opened: for reading, writing, appending, and possibly other modes. It also indexes into a third table called the inode table that describes the actual und

Evaluating and Interpreting Deep Convolutional Neural Networks via Non-negative Matrix Factorization

Edo Collins

With ever greater computational resources and more accessible software, deep neural networks have become ubiquitous across industry and academia. Their remarkable ability to generalize to new samples defies the conventional view, which holds that complex, over-parameterized networks would be prone to overfitting. This apparent discrepancy is exacerbated by our inability to inspect and interpret the high-dimensional, non-linear, latent representations they learn, which has led many to refer to neural networks as black-boxes''. The Law of Parsimony states that simpler solutions are more likely to be correct than complex ones''. Since they perform quite well in practice, a natural question to ask, then, is in what way are neural networks simple? We propose that compression is the answer. Since good generalization requires invariance to irrelevant variations in the input, it is necessary for a network to discard this irrelevant information. As a result, semantically similar samples are mapped to similar representations in neural network deep feature space, where they form simple, low-dimensional structures. Conversely, a network that overfits relies on memorizing individual samples. Such a network cannot discard information as easily. In this thesis we characterize the difference between such networks using the non-negative rank of activation matrices. Relying on the non-negativity of rectified-linear units, the non-negative rank is the smallest number that admits an exact non-negative matrix factorization. We derive an upper bound on the amount of memorization in terms of the non-negative rank, and show it is a natural complexity measure for rectified-linear units. With a focus on deep convolutional neural networks trained to perform object recognition, we show that the two non-negative factors derived from deep network layers decompose the information held therein in an interpretable way. The first of these factors provides heatmaps which highlight similarly encoded regions within an input image or image set. We find that these networks learn to detect semantic parts and form a hierarchy, such that parts are further broken down into sub-parts. We quantitatively evaluate the semantic quality of these heatmaps by using them to perform semantic co-segmentation and co-localization. In spite of the convolutional network we use being trained solely with image-level labels, we achieve results comparable or better than domain-specific state-of-the-art methods for these tasks. The second non-negative factor provides a bag-of-concepts representation for an image or image set. We use this representation to derive global image descriptors for images in a large collection. With these descriptors in hand, we perform two variations content-based image retrieval, i.e. reverse image search. Using information from one of the non-negative matrix factors we obtain descriptors which are suitable for finding semantically related images, i.e., belonging to the same semantic category as the query image. Combining information from both non-negative factors, however, yields descriptors that are suitable for finding other images of the specific instance depicted in the query image, where we again achieve state-of-the-art performance.

EPFL2019

Thermomechanical properties of honeycomb lattices from internal-coordinates potentials: the case of graphene and hexagonal boron nitride

Francesco Libbi, Nicola Marzari

Lattice dynamics in low-dimensional materials and, in particular, the quadratic behaviour of the flexural acoustic modes play a fundamental role in their thermomechanical properties. A first-principles evaluation of these can be very demanding, and can be affected by numerical noise that breaks translational or rotational invariance. In order to overcome these challenges, we study the Gartstein internal-coordinate potential and tune its 13 parameters on the first-principles interatomic force constants for graphene. We show that the resulting potential not only reproduces very well the phonon dispersions of graphene, but also those of carbon nanotubes of any diameter and chirality. The addition of a cubic term allows also to reproduce the dominant anharmonic terms, leading to a very good estimate of the lattice thermal conductivity. Finally, this potential form works very well also for boron nitride, provided it is fitted on the short-range (analytical) part of the interatomic force constants, and augmented thereafter with the long-range dielectric contribution. This consideration underscores how potentials based on short-ranged descriptors should be fit, in polar materials, to the short-range part of the first-principles interactions, and complemented by long-range analytical dielectric models parametrized on the same first-principles calculations.

IOP PUBLISHING LTD2021

A Comparison Between Joint Coordinate System and Attitude Vector for Multi-segment foot Kinematics

Kamiar Aminian, Xavier Crevoisier, Julien Favre, Hossein Rouhani

The joint angles of multi-segment foot models have been primarily described using two mathematical methods: the joint coordinate system and the attitude vector. This study aimed to determine whether the angles obtained through these two descriptors are comparable, and whether these descriptors have similar sensitivity to experimental errors. Six subjects walked eight times on an instrumented walkway while the joint angles among shank, hindfoot, medial forefoot, and lateral forefoot were measured. The angles obtained using both descriptors and their sensitivity to experimental errors were compared. There was no overall significant difference between the ranges of motion obtained using both descriptors. However, median differences of more than 6 were noticed for the medial-lateral forefoot joint. For all joints and rotation planes, both descriptors provided highly similar angle patterns (median correlation coefficient: R>0.90), except for the medial-lateral forefoot angle in the transverse plane (median R=0.77). The joint coordinate system was significantly more sensitive to anatomical landmarks misplacement errors. However, the absolute differences of sensitivity were small relative to the joints ranges of motion. In conclusion, the angles obtained using these two descriptors were not identical, but were similar for at least the shank-hindfoot and hindfoot-medial forefoot joints. Therefore, the angle comparison across descriptors is possible for these two joints. Comparison should be done more carefully for the medial-lateral forefoot joint. Moreover, despite different sensitivities to experimental errors, the effects of the experimental errors on the angles were small for both descriptors suggesting that both descriptors can be considered for multi-segment foot models. (C) 2012 Elsevier Ltd. All rights reserved.