SuperSFL: Resource-Heterogeneous Federated Split Learning with Weight-Sharing Supernet (Under review)
SuperSFL is a novel federated split learning framework that integrates supernet architectures to address resource and data heterogeneity in distributed machine learning systems. The framework employs dynamic subnetwork generation and dual-gradient backpropagation, enabling efficient training across heterogeneous devices while reducing server dependency through a client-side classifier. Experiments on CIFAR-10 and CIFAR-100 demonstrate SuperSFL's superior performance, achieving 20× faster convergence and 14× reduced training time compared to traditional approaches while maintaining higher accuracy and lower resource consumption in resource-constrained environments.
Bangla hate speech detection on social media using attention-based recurrent neural network.
De Gruyter - Journal of Intelligent Systems - 10.1515/jisys-2020-0060
Hate speech has spread more rapidly through the daily use of technology and, most notably, by sharing your opinions or feelings on social media in a negative aspect. Although numerous works have been carried out in detecting hate speeches in English, German, and other languages, very few works have been carried out in the context of the Bengali language. In contrast, millions of people communicate on social media in Bengali. The few existing works that have been carried out need improvements in both accuracy and interpretability.
This project proposed encoder–decoder-based machine learning model, a popular tool in NLP, to classify users' Bengali comments from Social media. In this article, the following contributions are made:
A Bangla Emot Module was created that helps to detect the emotions lying behind the emoji and emoticons. This helps to clarify the type of hate speech more clearly for Bangla Language.
Supervised machine learning algorithms such as attention LSTM- and Gated Recurrent Unit (GRU) based decoders was applied to the model. The model was then improved by using the Attention mechanism.
A real-time face mask detection and social distancing system for COVID-19 using Attention-InceptionV3 model
Journal of Engineering Advancements - 10.38032/jea.2022.01.001
One of the deadliest pandemics is now happening in the current world due to COVID-19. This contagious virus is spreading like wildfire around the whole world. To minimize the spreading of this virus, World Health Organization (WHO) has made protocols mandatory for wearing face masks and maintaining 6 feet physical distance.
We have applied some methods in deep learning which can detect the face mask and measure the social distance between people. We have summarized those in the following points -
We have used the object detection method in both cases.
Our system can detect in real-time video streams.
Comparing Machine Learning Algorithms (MLAs) to predict poverty incidences in Bangladesh (Ongoing)
The potential impact of this research is that to improve the country’s economic conditions as an entire by uplifting the poorest segments and giving them access to a stronger life. According to the industry-standard method of data mining, we carried out the following stages: data requirements, data acquisition, data preparation, data cleansing, exploratory data analysis, modeling, and algorithm. An efficient model is established by analyzing socio-economic conditions of household characteristics, the concerned can identify which households have the best need for financial aid assistance. By accurately predicting the household poverty level, we can assist with evaluating the social need in different administration levels of Bangladesh.
A Classification analysis of Cyberbullying Background (Researchgate)
Cyberbullying is a proceeding in which an individual or group harass or sabotages someone else on the internet. In this study, a personality and behavior analysis of a cyberbully has been conducted through a survey of 105 (63 males and 42 females) students from a university. Considering the feedback from the participants, it has been observed that in most cases, the victim thinks that the oppressors have psychological disorders, depression, loneliness, or family issue. As human beings, cyberbullies are overemotional, perverts or narcissists. They have lower self-esteem than others which is related to their marital status. Most of the perpetrators show their aggression by sending offensive messages or by threatening the victim in real life or virtual life. This research will raise awareness of the internet users of Bangladesh.