arXiv subjects

Sankha Subhra Mukherjee

Publications and source records attributed to Sankha Subhra Mukherjee.

7 recordsLinked to original sources

arXiv2022

ProSelfLC: Progressive Self Label Correction Towards A Low-Temperature Entropy State

There is a family of label modification approaches including self and non-self label correction (LC), and output regularisation. They are widely used for training robust deep neural networks (DNNs), but have not been mathematically and thoroughly analysed together. We study them and discover three key issues: (1) We are more interested in adopting Self LC as it leverages its own knowledge and requires no auxiliary models. However, it is unclear how to adaptively trust a learner as the training proceeds. (2) Some methods penalise while the others reward low-entropy (i.e., high-confidence) predictions, prompting us to ask which one is better. (3) Using the standard training setting, a learned model becomes less confident when severe noise exists. Self LC using high-entropy knowledge would generate high-entropy targets. To resolve the issue (1), inspired by a well-accepted finding, i.e., deep neural networks learn meaningful patterns before fitting noise, we propose a novel end-to-end method named ProSelfLC, which is designed according to the learning time and prediction entropy. Concretely, for any data point, we progressively and adaptively trust its predicted probability distribution versus its annotated one if a network has been trained for a relatively long time and the prediction is of low entropy. For the issue (2), the effectiveness of ProSelfLC defends entropy minimisation. By ProSelfLC, we empirically prove that it is more effective to redefine a semantic low-entropy state and optimise the learner toward it. To address the issue (3), we decrease the entropy of self knowledge using a low temperature before exploiting it to correct labels, so that the revised labels redefine low-entropy target probability distributions. We demonstrate the effectiveness of ProSelfLC through extensive experiments in both clean and noisy settings, and on both image and protein datasets.

cs.LG↗

arXiv2019

Texture Synthesis Guided Deep Hashing for Texture Image Retrieval

With the large-scale explosion of images and videos over the internet, efficient hashing methods have been developed to facilitate memory and time efficient retrieval of similar images. However, none of the existing works uses hashing to address texture image retrieval mostly because of the lack of sufficiently large texture image databases. Our work addresses this problem by developing a novel deep learning architecture that generates binary hash codes for input texture images. For this, we first pre-train a Texture Synthesis Network (TSN) which takes a texture patch as input and outputs an enlarged view of the texture by injecting newer texture content. Thus it signifies that the TSN encodes the learnt texture specific information in its intermediate layers. In the next stage, a second network gathers the multi-scale feature representations from the TSN's intermediate layers using channel-wise attention, combines them in a progressive manner to a dense continuous representation which is finally converted into a binary hash code with the help of individual and pairwise label information. The new enlarged texture patches also help in data augmentation to alleviate the problem of insufficient texture data and are used to train the second stage of the network. Experiments on three public texture image retrieval datasets indicate the superiority of our texture synthesis guided hashing approach over current state-of-the-art methods.

cs.CV↗

arXiv2014

Electrical detection of microwave assisted magnetization reversal by spin pumping

Microwave assisted magnetization reversal has been investigated in a bilayer system of Pt/ferromagnet by detecting a change in the polarity of the spin pumping signal. The reversal process is studied in two material systems, Pt/CoFeB and Pt/NiFe, for different aspect ratios. The onset of the switching behavior is indicated by a sharp transition in the spin pumping voltage. At a threshold value of the external field, the switching process changes from partial to full reversal with increasing microwave power. The proposed method provides a simple way to detect microwave assisted magnetization reversal.

cond-mat.mtrl-sci↗

arXiv2013

Characterization of magnetostatic surface spin waves in magnetic thin films: evaluation for microelectronic applications

The authors have investigated the possibility of utilizing spin waves for inter- and intra-chip communications, and as logic elements using both simulations and experimental techniques. Through simulations it has been shown that the decay lengths of magnetostatic spin waves are affected most by the damping parameter, and least by the exchange stiffness constant. The damping and dispersion properties of spin waves limit the attenuation length to several tens of microns. Thus, we have ruled out the possibility of inter-chip communications via spin waves. Experimental techniques for the extraction of the dispersion relationship have also been demonstrated, along with experimental demonstrations of spin wave interference for amplitude modulation. The effectiveness of spin wave modulation through interference, along with the capability of determining the spin wave dispersion relationships electrically during manufacturing and testing phase of chip production may pave the way for using spin waves in analog computing wherein the circuitry required for performing similar functionality becomes prohibitive.

cond-mat.mtrl-sci↗

arXiv2012

Spin waves interference from rising and falling edges of electrical pulses

The authors have investigated the effect of the electrical pulse width of input excitations on the generated spin waves in a NiFe strip using pulse inductive time domain measurements. The authors have shown that the spin waves resulting from the rising- and the falling-edges of input excitation pulses interfere either constructively or destructively, and have provided conditions for obtaining spin wave packets with maximum intensity at different bias conditions.

cond-mat.mtrl-sci↗

arXiv2012

Attenuation characteristics of spin pumping signal due to travelling spin waves

The authors have investigated the contribution of the surface spin waves to spin pumping. A Pt/NiFe bilayer has been used for measuring spin waves and spin pumping signals simultaneously. The theoretical framework of spin pumping resulting from ferromagnetic resonance has been extended to incorporate spin pumping due to spin waves. Equations for the effective area of spin pumping due to spin waves have been derived. The amplitude of the spin pumping signal resulting from travelling waves is shown to decrease more rapidly with precession frequency than that resulting from standing waves and show good agreement with the experimental data.

cond-mat.mtrl-sci↗

arXiv2012

Interference-mediated intensity modulation of spin waves

The modulation of propagating spin-wave amplitude in Ni81Fe19 (Py) films, resulting from constructive and destructive interference of spin wave, has been demonstrated. Spin waves were excited and detected inductively using pulse inductive time domain measurements. Two electrical impulses were used for launching two interfering Gaussian spin wave packets in Py films. The applied bias magnetic field or the separation between two pulses was used for tuning the amplitude of the resulting spin wave packets. This may thus be useful for spin wave based low-power information transfer and processing.

cond-mat.mtrl-sci↗