Harshil Sanghvi

Mr. Harshil Sanghvi

Nirma University, Ahmedabad, India
orcid

Research Focus & Metrics

Academic Career Length

5Years (2022-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
4 September 2022 in Mathematics
Verified
Journal Article
Open access
PDF file
DPDev Patel
Dev Patel
HSHarshil Sanghvi
Harshil Sanghvi
NJNilesh Kumar Jadav
Nilesh Kumar Jadav
RGRajesh Gupta
Rajesh Gupta
STSudeep Tanwar
Sudeep Tanwar
RSRavi Sharma
Ravi Sharma
Detecting malicious activity in advance has become increasingly important for public safety, economic stability, and national security. However, the disparity in living standards incites the minds of certain undesirable members of society to commit crimes, which may disrupt society’s stability and mental calm. Breakthroughs in deep learning (DL) make it feasible to address such challenges and construct a complete intelligent framework that automatically detects such malicious behaviors. Motivated by this, we propose a convolutional neural network (CNN)-based Xception model, i.e., BlockCrime, to detect crimes and improve public safety. Furthermore, we integrate blockchain technology to securely store the detected crime scene locations and alert the nearest law enforcement authorities. Due to the scarcity of the dataset, transfer learning has been preferred, in which a CNN-based Xception model is used. The redesigned Xception architecture is evaluated against various assessment measures, including accuracy, F1 score, precision, and recall, where it outperforms existing CNN architectures in terms of train accuracy, i.e., 96.57%.

26 July 2022
Verified
Book Chapter
Closed access
PDF file
PGPranshav Ritesh Gajjar
Pranshav Ritesh Gajjar
PSPooja Shah
Pooja Shah
HSHarshil Sanghvi
Harshil Sanghvi

16 October 2023
Verified
Conference Paper
Closed access
HSHarshil Sanghvi
Harshil Sanghvi
SMSapan H Mankad
Sapan H Mankad
Playback spoofing, the act of impersonating another speaker by playing back recordings of their voice, poses a significant threat tot he security o f Automatic Speaker Veri-fication (ASV) systems. This paper introduces anew method for detecting playback spoofing in A SV systems. O ur approach combines temporal and spectral features with Machine Learning (ML) and Deep Learning (DL) techniques to distinguish between playback spoofed and genuine speech. We employed Recursive Feature Elimination (RFE) to retain the most relevant features, improving performance. The results of our evaluation on a subset of ASV spoof 2019 dataset show that the combination of RFE and XGBoost produced outstanding results, with accuracy rising from 89.07% to 99.86% and Equal Error Rate (EER) decreasing from 48.05% to a mere 0.69%. Our study highlights the significance of RFE in enhancing the performance of speaker verification systems and the potential of XGBoost as an algorithm for detecting playback spoofing attacks.

3 March 2026 in bioRxiv
Verified
Preprint
Open access
PDF file
AGAndrew Vincent Gallagher
Andrew Vincent Gallagher
AWAmber Victoria Wilson
Amber Victoria Wilson
SLSaheed Lawal
Saheed Lawal
HSHarshil Sanghvi
Harshil Sanghvi
JIJaved Iqbal
Javed Iqbal
PSPrerana Shrestha
Prerana Shrestha
Animals must flexibly discriminate between threat and non-threat to deploy adaptive defensive strategies. We introduce a single-episode differential signaled active avoidance (DSAA) paradigm that temporally dissociates acquisition, consolidation, and retrieval of instrumental avoidance memory. Interleaving a behaviorally noncontingent neutral cue (CS-) with a threat-predictive, behaviorally contingent cue (CS+) enhanced long-term memory without altering acquisition, demonstrating that differential contingency structure selectively reinforces memory consolidation rather than influencing learning performance. Naïve animals inferred contingencies within a single episode, and discrimination achieved during training predicted retrieval precision. However, discrimination operated within defined boundary conditions: overtraining or elevated threat intensity destabilized cue specificity and promoted persistent avoidance generalization. Under high-threat conditions, freezing and shuttling co-emerged as complementary defensive responses, indicating a shift from precise cue-based encoding to a generalized defensive state. Remote retrieval recruited oxytocin receptor-expressing cells in the medial prefrontal cortex and activated mTORC1-dependent translational signaling, implicating protein synthesis in maintenance of discriminative avoidance memory. In a Tuberous Sclerosis Complex model with Tsc2 haploinsufficiency in oxytocin-responsive cells, males displayed intact acquisition but generalized avoidance at both recent and remote time points, a deficit not rescued by additional training. These findings identify oxytocin-modulated translational control as a molecular gate stabilizing threat-safety discrimination and show that disruption of this axis - by excessive threat or reduced Tsc2 gene dosage - biases memory toward pathological generalization, providing a mechanistic framework for safety-learning deficits in neurodevelopmental and anxiety-related disorders.