Source author record

Ali Akbari

Ali Akbari appears in the imported research catalog. Authorship, coauthor and topic links are available while profile ownership is still unclaimed.

ResearcherUnclaimed source record

Computer Vision Machine Learning cond-mat.mtrl-sci

Catalog footprint

What is connected

3works

3topics

4close collaborators

Actions

Connect this record

Open graph Browse works

Inspect adjacent papers, topics, institutions and collaborators without losing the researcher page.

Building this map preview

BZPEER is loading the nearby papers, people, topics and institutions for this page.

preprint2021arXiv

NPT-Loss: A Metric Loss with Implicit Mining for Face Recognition

Face recognition (FR) using deep convolutional neural networks (DCNNs) has seen remarkable success in recent years. One key ingredient of DCNN-based FR is the appropriate design of a loss function that ensures discrimination between various identities. The state-of-the-art (SOTA) solutions utilise normalised Softmax loss with additive and/or multiplicative margins. Despite being popular, these Softmax+margin based losses are not theoretically motivated and the effectiveness of a margin is justified only intuitively. In this work, we utilise an alternative framework that offers a more direct mechanism of achieving discrimination among the features of various identities. We propose a novel loss that is equivalent to a triplet loss with proxies and an implicit mechanism of hard-negative mining. We give theoretical justification that minimising the proposed loss ensures a minimum separability between all identities. The proposed loss is simple to implement and does not require heavy hyper-parameter tuning as in the SOTA solutions. We give empirical evidence that despite its simplicity, the proposed loss consistently achieves SOTA performance in various benchmarks for both high-resolution and low-resolution FR tasks.

preprint2020arXiv

A Convolutional Baseline for Person Re-Identification Using Vision and Language Descriptions

Classical person re-identification approaches assume that a person of interest has appeared across different cameras and can be queried by one of the existing images. However, in real-world surveillance scenarios, frequently no visual information will be available about the queried person. In such scenarios, a natural language description of the person by a witness will provide the only source of information for retrieval. In this work, person re-identification using both vision and language information is addressed under all possible gallery and query scenarios. A two stream deep convolutional neural network framework supervised by cross entropy loss is presented. The weights connecting the second last layer to the last layer with class probabilities, i.e., logits of softmax layer are shared in both networks. Canonical Correlation Analysis is performed to enhance the correlation between the two modalities in a joint latent embedding space. To investigate the benefits of the proposed approach, a new testing protocol under a multi modal ReID setting is proposed for the test split of the CUHK-PEDES and CUHK-SYSU benchmarks. The experimental results verify the merits of the proposed system. The learnt visual representations are more robust and perform 22\% better during retrieval as compared to a single modality system. The retrieval with a multi modal query greatly enhances the re-identification capability of the system quantitatively as well as qualitatively.

preprint2012arXiv

Challenges in Truncating the Hierarchy of Time-Dependent Reduced Density Matrices Equations: Open Problems

In this work, we analyze the Born, Bogoliubov, Green, Kirkwood and Yvon (BBGKY) hierarchy of equations for describing the full time-evolution of a many-body fermionic system in terms of its reduced density matrices (at all orders). We provide an exhaustive study of the challenges and open problems linked to the truncation of such hierarchy of equations to make them practically applicable. We restrict our analysis to the coupled evolution of the one- and two-body reduced density matrices, where higher order correlation effects are embodied into the approximation used to close the equations. We prove that within this approach, the number of electrons and total energy are conserved, regardless of the employed approximation. Further, we demonstrate that although most of the truncation schemes available in the literature give acceptable ground state energy, when applied to describe driven electron dynamics exhibit undesirable and unphysical behavior, e.g., violation and even divergence in local electronic density, both in weakly- and strongly-correlated regimes. We illustrate and analyze these problems within the few-site Hubbard model. The model can be solved exactly and provides a unique reference for our detailed study of electron dynamics for different values of interaction, different initial conditions, and large set of approximations considered here. Moreover, we study the role of compatibility between two hierarchical equations, and positive- semidefiniteness of reduced density matrices in instability of electron dynamics. We show that even if the used approximation holds the compatibility, electron dynamics can still diverge when positive- definitiveness is violated. We propose some partial solutions of such problem and point the main paths for future work in order to make this approach applicable for the description of the correlated electron dynamics in complex systems.