Searcharxiv⌕ Search

arXiv subjects

Simant Dube

Publications and source records attributed to Simant Dube.

5 recordsLinked to original sources
arXiv2021

Expressive Power and Loss Surfaces of Deep Learning Models

The goals of this paper are two-fold. The first goal is to serve as an expository tutorial on the working of deep learning models which emphasizes geometrical intuition about the reasons for success of deep learning. The second goal is to complement the current results on the expressive power of deep learning models and their loss surfaces with novel insights and results. In particular, we describe how deep neural networks carve out manifolds especially when the multiplication neurons are introduced. Multiplication is used in dot products and the attention mechanism and it is employed in capsule networks and self-attention based transformers. We also describe how random polynomial, random matrix, spin glass and computational complexity perspectives on the loss surfaces are interconnected.

cs.LG↗
arXiv2018

High Dimensional Spaces, Deep Learning and Adversarial Examples

In this paper, we analyze deep learning from a mathematical point of view and derive several novel results. The results are based on intriguing mathematical properties of high dimensional spaces. We first look at perturbation based adversarial examples and show how they can be understood using topological and geometrical arguments in high dimensions. We point out mistake in an argument presented in prior published literature, and we present a more rigorous, general and correct mathematical result to explain adversarial examples in terms of topology of image manifolds. Second, we look at optimization landscapes of deep neural networks and examine the number of saddle points relative to that of local minima. Third, we show how multiresolution nature of images explains perturbation based adversarial examples in form of a stronger result. Our results state that expectation of $L_2$-norm of adversarial perturbations is $O\left(\frac{1}{\sqrt{n}}\right)$ and therefore shrinks to 0 as image resolution $n$ becomes arbitrarily large. Finally, by incorporating the parts-whole manifold learning hypothesis for natural images, we investigate the working of deep neural networks and root causes of adversarial examples and discuss how future improvements can be made and how adversarial examples can be eliminated.

cs.CV↗
arXiv2009

An Iterative Fingerprint Enhancement Algorithm Based on Accurate Determination of Orientation Flow

We describe an algorithm to enhance and binarize a fingerprint image. The algorithm is based on accurate determination of orientation flow of the ridges of the fingerprint image by computing variance of the neighborhood pixels around a pixel in different directions. We show that an iterative algorithm which captures the mutual interdependence of orientation flow computation, enhancement and binarization gives very good results on poor quality images.

cs.CV↗
arXiv2008

Computation of Maximal Resolution of Copy Number Variation on a Nanofluidic Device using Digital PCR

Copy Number Variations (CNVs) of regions of the human genome are important in disease association studies.The digital array is a nanofluidic biochip which utilizes integrated channels and valves that partition mixtures of sample and reagents into 765 nanovolume reaction chambers. It was recently shown how one can perform statistical analysis of CNV in a DNA sample the digital array. In particular, it was shown how one can accurately estimate the true concentration of the molecules in the DNA sample and then determine the ratios of different sequences along with statistical confidence intervals on these estimations. In this paper we perform computation of maximum number of copies which can be distinguished using the digital array which gives its resolution in terms of its ability to determine CNV. Then, we demonstrate the usefulness of the mathematical analysis to solve an important real-world problem of determination of the copy number of X chromosome as our example application.

q-bio.GN↗