arXiv subjects

Shilpak Banerjee

Publications and source records attributed to Shilpak Banerjee.

14 recordsLinked to original sources

arXiv2024

Weak mixing behavior for the projectivized derivative extension

In both smooth and analytic categories, we construct examples of diffeomorphisms of topological entropy zero with intricate ergodic properties. On any smooth compact connected manifold of dimension 2 admitting a nontrivial circle action, we construct a smooth diffeomorphism whose differential is weakly mixing with respect to a smooth measure in the projectivization of the tangent bundle. In case of the 2-torus, we also obtain the analytic counterpart of such a diffeomorphism. The constructions are based on a quantitative version of the ``Approximation by Conjugation'' method, which involves explicitly defined conjugation maps, partial partitions, and the adaptation of a specific analytic approximation technique.

math.DS↗

arXiv2023

Ergodic and mixing quantum channels: From two-qubit to many-body quantum systems

The development of classical ergodic theory has had a significant impact in the areas of mathematics, physics, and, in general, applied sciences. The quantum ergodic theory of Hamiltonian dynamics has its motivations to understand thermodynamics and statistical mechanics. Quantum channel, a completely positive trace-preserving map, represents a most general representation of quantum dynamics and is an essential aspect of quantum information theory and quantum computation. In this work, we study the ergodic theory of quantum channels by characterizing different levels of ergodic hierarchy from integrable to mixing. The quantum channels on single systems are constructed from the unitary operators acting on bipartite states and tracing out the environment. The interaction strength of these unitary operators measured in terms of operator entanglement provides sufficient conditions for the channel to be mixing. By using block diagonal unitary operators, we construct a set of non-ergodic channels. By using canonical form of two-qubit unitary operator, we analytically construct the channels on single qubit ranging from integrable to mixing. Moreover, we also study interacting many-body quantum systems that include the famous Sachdev-Ye-Kitaev (SYK) model and show that they display mixing within the framework of the quantum channel.

quant-ph↗

arXiv2021

SMU: smooth activation function for deep networks using smoothing maximum technique

Deep learning researchers have a keen interest in proposing two new novel activation functions which can boost network performance. A good choice of activation function can have significant consequences in improving network performance. A handcrafted activation is the most common choice in neural network models. ReLU is the most common choice in the deep learning community due to its simplicity though ReLU has some serious drawbacks. In this paper, we have proposed a new novel activation function based on approximation of known activation functions like Leaky ReLU, and we call this function Smooth Maximum Unit (SMU). Replacing ReLU by SMU, we have got 6.22% improvement in the CIFAR100 dataset with the ShuffleNet V2 model.

cs.LG↗

arXiv2021

SAU: Smooth activation function using convolution with approximate identities

Well-known activation functions like ReLU or Leaky ReLU are non-differentiable at the origin. Over the years, many smooth approximations of ReLU have been proposed using various smoothing techniques. We propose new smooth approximations of a non-differentiable activation function by convolving it with approximate identities. In particular, we present smooth approximations of Leaky ReLU and show that they outperform several well-known activation functions in various datasets and models. We call this function Smooth Activation Unit (SAU). Replacing ReLU by SAU, we get 5.12% improvement with ShuffleNet V2 (2.0x) model on CIFAR100 dataset.

cs.LG↗

arXiv2021

Slow entropy for some Anosov-Katok diffeomorphisms

The Anosov-Katok method is one of the most powerful tools of constructing smooth volume-preserving diffeomorphisms of entropy zero with prescribed ergodic or topological properties. To measure the complexity of systems with entropy zero, invariants like slow entropy have been introduced. In this article we develop several mechanisms facilitating computation of topological and measure-theoretic slow entropy of Anosov-Katok diffeomorphisms.

math.DS↗

arXiv2021

ErfAct and Pserf: Non-monotonic Smooth Trainable Activation Functions

An activation function is a crucial component of a neural network that introduces non-linearity in the network. The state-of-the-art performance of a neural network depends also on the perfect choice of an activation function. We propose two novel non-monotonic smooth trainable activation functions, called ErfAct and Pserf. Experiments suggest that the proposed functions improve the network performance significantly compared to the widely used activations like ReLU, Swish, and Mish. Replacing ReLU by ErfAct and Pserf, we have 5.68% and 5.42% improvement for top-1 accuracy on Shufflenet V2 (2.0x) network in CIFAR100 dataset, 2.11% and 1.96% improvement for top-1 accuracy on Shufflenet V2 (2.0x) network in CIFAR10 dataset, 1.0%, and 1.0% improvement on mean average precision (mAP) on SSD300 model in Pascal VOC dataset.

cs.NE↗

arXiv2021

Orthogonal-Pad\'e Activation Functions: Trainable Activation functions for smooth and faster convergence in deep networks

We have proposed orthogonal-Pad\'e activation functions, which are trainable activation functions and show that they have faster learning capability and improves the accuracy in standard deep learning datasets and models. Based on our experiments, we have found two best candidates out of six orthogonal-Pad\'e activations, which we call safe Hermite-Pade (HP) activation functions, namely HP-1 and HP-2. When compared to ReLU, HP-1 and HP-2 has an increment in top-1 accuracy by 5.06% and 4.63% respectively in PreActResNet-34, by 3.02% and 2.75% respectively in MobileNet V2 model on CIFAR100 dataset while on CIFAR10 dataset top-1 accuracy increases by 2.02% and 1.78% respectively in PreActResNet-34, by 2.24% and 2.06% respectively in LeNet, by 2.15% and 2.03% respectively in Efficientnet B0.

cs.NE↗

arXiv2020

Slow entropy of some combinatorial constructions

Measure-theoretic slow entropy is a more refined invariant than the classical measure-theoretic entropy to characterize the complexity of dynamical systems with subexponential growth rates of distinguishable orbit types. In this paper we prove flexibility results for the values of upper and lower polynomial slow entropy of rigid transformations as well as maps admitting a good cyclic approximation. Moreover, we show that there cannot exist a general upper bound on the lower measure-theoretic slow entropy for systems of finite rank.

math.DS↗

arXiv2020

EIS -- a family of activation functions combining Exponential, ISRU, and Softplus

Activation functions play a pivotal role in the function learning using neural networks. The non-linearity in the learned function is achieved by repeated use of the activation function. Over the years, numerous activation functions have been proposed to improve accuracy in several tasks. Basic functions like ReLU, Sigmoid, Tanh, or Softplus have been favorite among the deep learning community because of their simplicity. In recent years, several novel activation functions arising from these basic functions have been proposed, which have improved accuracy in some challenging datasets. We propose a five hyper-parameters family of activation functions, namely EIS, defined as, \[ \frac{x(\ln(1+e^x))^α}{\sqrt{β+γx^2}+δe^{-θx}}. \] We show examples of activation functions from the EIS family which outperform widely used activation functions on some well known datasets and models. For example, $\frac{x\ln(1+e^x)}{x+1.16e^{-x}}$ beats ReLU by 0.89\% in DenseNet-169, 0.24\% in Inception V3 in CIFAR100 dataset while 1.13\% in Inception V3, 0.13\% in DenseNet-169, 0.94\% in SimpleNet model in CIFAR10 dataset. Also, $\frac{x\ln(1+e^x)}{\sqrt{1+x^2}}$ beats ReLU by 1.68\% in DenseNet-169, 0.30\% in Inception V3 in CIFAR100 dataset while 1.0\% in Inception V3, 0.15\% in DenseNet-169, 1.13\% in SimpleNet model in CIFAR10 dataset.

cs.LG↗

arXiv2020

TanhSoft -- a family of activation functions combining Tanh and Softplus

Deep learning at its core, contains functions that are composition of a linear transformation with a non-linear function known as activation function. In past few years, there is an increasing interest in construction of novel activation functions resulting in better learning. In this work, we propose a family of novel activation functions, namely TanhSoft, with four undetermined hyper-parameters of the form tanh(αx+βe^{γx})ln(δ+e^x) and tune these hyper-parameters to obtain activation functions which are shown to outperform several well known activation functions. For instance, replacing ReLU with xtanh(0.6e^x)improves top-1 classification accuracy on CIFAR-10 by 0.46% for DenseNet-169 and 0.7% for Inception-v3 while with tanh(0.87x)ln(1 +e^x) top-1 classification accuracy on CIFAR-100 improves by 1.24% for DenseNet-169 and 2.57% for SimpleNet model.

cs.NE↗

arXiv2020

Real-analytic realization of Uniform Circular Systems and some applications

Recently Matthew Foreman and Benjamin Weiss showed in a series of papers that smooth ergodic diffeomorphisms of a compact manifold are unclassifiable up to measure-isomorphism. In this paper we show that the uniform circular systems used in the work of Foreman-Weiss admit real-analytic realizations on the torus. As a consequence we obtain the same anti-classification result for real-analytic ergodic diffeomorphisms on the torus. In another application we show the existence of an uncountable family of pairwise non-Kakutani equivalent real-analytic diffeomorphisms on the torus.

math.DS↗

arXiv2017

Real-analytic diffeomorphisms with homogeneous spectrum and disjointness of convolutions

On any torus $\mathbb{T}^d$, $d \geq 2$, we prove the existence of a real-analytic diffeomorphism $T$ with a good approximation of type $\left(h,h+1\right)$, a maximal spectral type disjoint with its convolutions and a homogeneous spectrum of multiplicity two for the Cartesian square $T\times T$. The proof is based on a real-analytic version of the Approximation by Conjugation-method.

math.DS↗

arXiv2016

Real-analytic AbC constructions on the torus

In this article we demonstrate a way to extend the AbC (approximation by conjugation) method invented by Anosov and Katok from the smooth category to the category of real-analytic diffeomorphisms on the torus. We present a general framework for such constructions and prove several results. In particular, we construct minimal but not uniquely ergodic diffeomorphisms and nonstandard real-analytic realizations of toral translations.

math.DS↗

arXiv2015

Non-standard real-analytic realizations of some rotations of the circle

We extend some aspects of the smooth approximation by conjugation method to the real-analytic set-up and create examples of zero entropy, uniquely ergodic real-analytic diffeomorphisms of the two dimensional torus metrically isomorphic to some (Liouvillian) irrational rotations of the circle.

math.DS↗