With 8+ years of experience in AI and data science, I’m passionate about solving complex problems at the intersection of Machine Learning, Deep Learning, Geometric Deep Learning, Explainable AI, and Causal AI. My expertise spans across Multivariate Statistical Analysis, Time Series Modeling, Stochastic Processes, Computer Vision, and NLP Applications.
I firmly believe that quality data always trumps sheer volume, and with the right collaboration between AI and domain experts, any data challenge can be tackled effectively. Whether it's building intelligent AI systems, optimizing machine learning models, or uncovering hidden insights, I thrive on turning data into meaningful solutions.
Beyond work, I’m a huge fan of books, cricket, movies, and TV series. But if you ask me my favorite way to unwind? It’s always music 🎶.
Let’s connect and explore the endless possibilities of AI together! 🚀
April 2022 - Present | Singapore
November 2021 - April 2022 | Dhaka
March 2021 - September 2021 | Dhaka
February 2020 - September 2021 | Japan
March 2019 - January 2020 | Dhaka
February 2018 - February 2019 | Dhaka
February 2017 - December 2017 | Dhaka
August 2024 | Organized by AI for Global Goals, supported by CIFAR, Oxford, LSE, a 3-day summer school focused on autonomous problem-solving agents.
July 2024 | Attended 3-day summer school by AI for Global Goals, CIFAR, Oxford on representation learning and generative AI applications.
July 2024 | Participated in a 3-day summer school organized by AI for Global Goals, CIFAR, Oxford, focused on the latest ML advancements, including deep learning and generative AI, for improving healthcare and biotechnology.
June 2023 - September 2023 | Co-designed by the Institution of Business Administration, University of Dhaka, Indian Institution of Management, Ahmedabad (IIMA), Indian Institution of Technology, Delhi (IITD), offered by the Management Development Program (MDP), IBA.
June 2020 - September 2020 | A learning journey for 10 IT & ITES companies in Bangladesh. The product development phase was actively mentored by NUS-ISS faculty members.
January 2019 - July 2019 | Specialized training in developing custom intellectual property to drive innovation across industries using state-of-the-art machine learning and AI.
LLM reasoning for solving riddles leverages the model's deep understanding of language, patterns, and logic. By analyzing key elements such as wordplay, metaphors, and ambiguous phrasing, the model identifies and interprets hidden clues. Using contextual knowledge and lateral thinking, the LLM generates accurate solutions. This process combines syntactic and semantic analysis to unlock meanings and relationships, making it a powerful tool for tackling complex riddles and puzzles.
I explore how Large Language Models (LLMs) can transform computer science education and impact the future of society by providing intelligent tutoring systems and AI-driven learning experiences.
This research investigates the capacity of Large Language Models (LLMs) to comprehend and solve Bengali folklore riddles, which are rich in wordplay and cultural nuances. The study aims to compile a comprehensive dataset of traditional Bengali riddles, fine-tune reasoning models, and assess their performance in understanding and solving these challenges. By enhancing AI’s ability to reason within specific cultural and linguistic contexts, this work makes significant contributions to the fields of cultural NLP, AI reasoning, and low-resource language processing.
This project develops a domain-adapted Bengali text summarization model using large language models (LLMs) like flanT5 and mT5. It fine-tunes these models on general and domain-specific datasets, including state, international, and sports categories, to improve accuracy in generating summaries. Trained on the XLSUM Bengali dataset, the project also creates a custom annotated dataset for domain adaptation. The findings highlight the effectiveness of the flanT5 XLSUM model in generating high-quality Bengali summaries for the state domain, advancing summarization tools for low-resource languages.
Join me in exploring cutting-edge research in AI, ML, and Software Engineering. Check out my research insights for more exciting opportunities!
April 2021 - August 2021
August 2020 - January 2021
August 2020 - January 2021
June 2020 - October 2020
January 2020 - June 2020
MM Akash, Rahul Deb Mohalder, Md Al Mamun Khan, Laboni Paul, Ferdous Bin Ali
2nd International Conference on Big Data, IoT and Machine Learning,
BIM(2023)
TLDR: Yoga has become essential for health, but human pose estimation, particularly with the complex Yoga-82 dataset, poses challenges. By fine-tuning VGG-16, ResNet, and DenseNet-121 models and applying Neural Architecture Search, DenseNet-121 achieved the best results with 85% top-1 accuracy and 96% top-5 accuracy, outperforming the current state-of-the-art.
Abstract:
Yoga has recently become an essential aspect of human existence for maintaining a healthy body and mind. People find it tough to devote time to the gym for workouts as their lives get more hectic and they work from home. This kind of human pose estimation is one of the notable problems as it has to deal with locating body key points or joints. Yoga-82, a benchmark dataset for large-scale yoga pose recognition with 82 classes, has challenging positions that could make precise annotations impossible. We have used VGG-16, ResNet-50, ResNet-101, and DenseNet-121 and finetuned them in different ways to get better results. We also used Neural Architecture Search to add more layers on top of this pre-trained architecture. The experimental result shows the best performance of DenseNet-121 having the top-1 accuracy of 85% and top-5 accuracy of 96% outperforming the current state-of-the-art result.
Rajana Akter, Shahnure Rabib, Rahul Deb Mohalder, Laboni Paul, Ferdous Bin Ali
2nd International Conference on Big Data, IoT and Machine Learning,
BIM(2023)
TLDR: The study introduces SCGNet, a deep learning architecture for intrusion detection, achieving 99.76% accuracy in network attack detection and 98.92% in attack type classification on the NSL-KDD dataset. The proposed system addresses the limitations of traditional IDSs in detecting complex and low-frequency attacks. A general data preprocessing pipeline is also introduced, applicable to similar datasets, alongside comparisons with conventional machine-learning methods.
Abstract:
Nawal Ayesha Khan, Sarah Jasim, Intisar Tahmid Naheen, Ferdous Bin Ali
International Conference on Deep Learning, Artificial Intelligence and Robotics 2023
TLDR:This paper develops a smart music recommendation system that uses both demographic data and psychological factors, like mood and personality, to suggest music genres. Ensemble algorithms performed best for "Happy" and "Gloomy" moods, while semi-supervised algorithms excelled for "Stressed" and "Relaxed" moods.
Abstract:
Individual music preferences can generally depend on severalfactors, namely demographic data such as gender, age, etc., and morespecific psychological factors like mood and personality. However, whilemainstream music applications tend to consider the former, they usuallyignore psychological factors that would otherwise allow for more accu-rate recommendations. In this paper, we attempt to develop a smartmusic recommendation system by comparing various machine-learningalgorithms to recommend a genre based on the user’s current mood,which is determined using a psychological scale, along with their basicdemographic data. A custom dataset was built, where the TIPI scalewas used to identify users’ personality types, which we then fed into themodels to generate recommendations. Based on the results, ensemble al-gorithms performed the best for the moods of “Happy” and “Gloomy”,while semi-supervised algorithms achieved the best scores for the moodsof “Stressed” and “Relaxed (PDF) Music Recommendation System Using Psychological Scale.
Ferdous Bin Ali, Rahul Deb Mohalder, Rajana Akter, Laboni Paul, Md Al Mamun Khan, Nurzahan Akter Joly
26th International Conference on Computer and Information Technology (ICCIT) 2023
TLDR: This research utilizes transfer learning with ResNet-50v2 and DenseNet-201 CNN architectures to detect potato diseases, particularly early and late blight. The study achieved state-of-the-art results across three different datasets, showcasing the potential for automated disease detection in potato crops.
Abstract:
Potato, a globally significant food crop ranking as
the fourth largest by production, is cultivated in various regions
worldwide. However, potato crops are notably susceptible to
fungal infections, leading to the occurrence of early blight and
late blight diseases. Timely disease control and management
measures are pivotal in augmenting crop yields and mitigating
agricultural losses for farmers. The capacity to automatically
discern diseased crops holds substantial promise for farmers.
Consequently, this research endeavors to present the power of
transfer learning in SoTA Convolutional Neural Network (CNN)
architecture ResNet-50v2 and DenseNet-201 which are finetuned
to the task of potato disease detection. In our experimental
endeavors, we employed three distinct datasets, and in each
instance, we attained state-of-the-art results.