Papers by Chandresh Pravin
arXiv (Cornell University), Dec 14, 2023
We propose a systematic analysis of deep neural networks (DNNs) based on a signal processing tech... more We propose a systematic analysis of deep neural networks (DNNs) based on a signal processing technique for network parameter removal, in the form of synaptic filters that identifies the fragility, robustness and antifragility characteristics of DNN parameters. Our proposed analysis investigates if the DNN performance is impacted negatively, invariantly, or positively on both clean and adversarially perturbed test datasets when the DNN undergoes synaptic filtering. We define three filtering scores for quantifying the fragility, robustness and antifragility characteristics of DNN parameters based on the performances for (i) clean dataset, (ii) adversarial dataset, and (iii) the difference in performances of clean and adversarial datasets. We validate the proposed systematic analysis on ResNet-18, ResNet-50, SqueezeNet-v1.1 and ShuffleNet V2 x1.0 network architectures for MNIST, CIFAR10 and Tiny ImageNet datasets. The filtering scores, for a given network architecture, identify network parameters that are invariant in characteristics across different datasets over learning epochs. Vice-versa, for a given dataset, the filtering scores identify the parameters that are invariant in characteristics across different network architectures. We show that our synaptic filtering method improves the test accuracy of ResNet and ShuffleNet models on adversarial datasets when only the robust and antifragile parameters are selectively retrained at any given epoch, thus demonstrating applications of the proposed strategy in improving model robustness.
Song Preference CLassification Dataset for Gen Z
This dataset contains audio recordings of 12 different accents across the UK: Northern Ireland, S... more This dataset contains audio recordings of 12 different accents across the UK: Northern Ireland, Scotland, Wales, North East England, North West England, Yorkshire and Humber, East Midlands, West Midlands, East of England, Greater London, South East England, South West England. We split the data into a Male: Female ratio of 1:1. The audio dataset was compiled using opensource YouTube videos and it a collation of different accents, the audio files were trimmed for uniformity. The Audio files are of length 30 seconds, with the first 5 seconds and last 5 seconds of the signal being blank. We also resample the audio signals at 8 kHz, again for uniformity and to remove any noise present in the audio signals whilst retaining the underlying characteristics. The intended application of this dataset was to be used in conjunction with a deep neural network for accent and gender classification tasks. This dataset was recorded for an experimentation looking into applying machine learning techniq...
Accent Classification Dataset
This dataset contains audio recordings of 12 different accents across the UK: Northern Ireland (N... more This dataset contains audio recordings of 12 different accents across the UK: Northern Ireland (NI), Scotland, Wales (SW), North East England (NE), North West England (NW), Yorkshire and Humber (YAH), East Midlands (EM), West Midlands (WM), East of England (EE), Greater London (GL), South East England (SE), South West England (SW). We split the data into a Male: Female ratio of 1:1, this is labelled with either '_M' for male or '_F' for female within the dataset. The audio dataset was compiled using opensource YouTube videos and it a collation of different accents, the audio files were trimmed for uniformity. The Audio files are of length 30 seconds, with the first 5 seconds and last 5 seconds of the signal being blank. We also resample the audio signals at 8 kHz, again for uniformity and to remove any noise present in the audio signals whilst retaining the underlying characteristics. The intended application of this dataset was to be used in conjunction with a deep ...
Advances in Intelligent Systems and Computing, 2021
This paper proposes a methodology for investigating musical preferences of the age group between ... more This paper proposes a methodology for investigating musical preferences of the age group between 18 and 24. We conducted an electroencephalogram (EEG) experiment to collect individual's responses to audio stimuli along with a measure of like or dislike for a piece of music. Machine learning (multilayer perceptron and support vector machine) classifiers and signal processing [independent component analysis (ICA)] techniques were applied on the pre-processed dataset of 10 participant's EEG signals and preference ratings. Our classification model classified song preference with high accuracy. The ICA based EEG signal processing enabled the identification of perceptual patterns via analysis of the spectral peaks which suggest that the recorded brain activities were dependent on the respective song's rating.
Proceedings of the Fourteenth Workshop on Semantic Evaluation, 2020
SemEval Task 4 Commonsense Validation and Explanation Challenge is to validate whether a system c... more SemEval Task 4 Commonsense Validation and Explanation Challenge is to validate whether a system can differentiate natural language statements that make sense from those that do not make sense. Two subtasks, A and B, are focused in this work, i.e., detecting against-common-sense statements and selecting explanations of why they are false from the given options. Intuitively, commonsense validation requires additional knowledge beyond the given statements. Therefore, we propose a system utilising pre-trained sentence transformer models based on BERT, RoBERTa and DistillBERT architectures to embed the statements before classification. According to the results, these embeddings can improve the performance of the typical MLP and LSTM classifiers as downstream models of both subtasks compared to regular tokenised statements. These embedded statements are shown to comprise additional information from external resources which help validate common sense in natural language.
Accent and Gender Recognition from English Language Speech and Audio Using Signal Processing and Deep Learning
This research is concerned with taking user input in the form of speech data to classify and then... more This research is concerned with taking user input in the form of speech data to classify and then predict which region of the United Kingdom the user is from and their gender. This research was conducted on regional accents, data preprocessing, Fourier transforms, and deep learning modeling. Due to the lack of publicly available datasets for this type of research, a dataset was created from scratch (12 regions with a 1:1 gender ratio). In this paper, we propose modeling the human’s voice accent and voice gender recognition as a classification task. We used a deep convolution neural network, and experimentally developed an architecture that maximized the classification accuracy of the mentioned tasks simultaneously. We also tested the model on publicly available spoken digit detests. We find that the gender classification is relatively easier to predict with high accuracy than the accent in our proposed multi-class classification model. Accent classification was found difficult because of the regional accent’s overlapping that prevents it from being classified with high accuracy
2020 19th IEEE International Conference on Machine Learning and Applications (ICMLA), 2020
We propose a novel deep learning based denoising filter selection algorithm for noisy Electrocard... more We propose a novel deep learning based denoising filter selection algorithm for noisy Electrocardiograph (ECG) signal preprocessing. ECG signals measured under clinical conditions, such as those acquired using skin contact devices in hospitals, often contain baseline signal disturbances and unwanted artefacts; indeed for signals obtained outside of a clinical environment, such as heart rate signatures recorded using noncontact radar systems, the measurements contain greater levels of noise than those acquired under clinical conditions. In this paper we focus on heart rate signals acquired using non-contact radar systems for use in assisted living environments. Such signals contain more noise than those measured under clinical conditions, and thus require a novel signal noise removal method capable of adaptive determining filters. Currently the most common method of removing noise from such a waveform is through the use of filters; the most popular filtering method amongst which is the wavelet filter. There are, however, circumstances in which using a different filtering method may result in higher signal-to-noiseratios (SNR) for a waveform; in this paper, we investigate the wavelet and elliptical filtering methods for the task of reducing noise in ECG signals acquired using assistive technologies. Our proposed convolutional neural network architecture classifies (with 92.8% accuracy) the optimum filtering method for noisy signal based on its expected SNR value.
Adversarial Robustness in Deep Learning: Attacks on Fragile Neurons
Lecture Notes in Computer Science
We identify fragile and robust neurons of deep learning architectures using nodal dropouts of the... more We identify fragile and robust neurons of deep learning architectures using nodal dropouts of the first convolutional layer. Using an adversarial targeting algorithm, we correlate these neurons with the distribution of adversarial attacks on the network. Adversarial robustness of neural networks has gained significant attention in recent times and highlights an intrinsic weaknesses of deep learning networks against carefully constructed distortion applied to input images. In this paper, we evaluate the robustness of state-of-the-art image classification models trained on the MNIST and CIFAR10 datasets against the fast gradient sign method attack, a simple yet effective method of deceiving neural networks. Our method identifies the specific neurons of a network that are most affected by the adversarial attack being applied. We, therefore, propose to make fragile neurons more robust against these attacks by compressing features within robust neurons and amplifying the fragile neurons proportionally.
Uploads
Papers by Chandresh Pravin