Anuran Bhattacharya[cite: 7]
I am an M.Tech. graduate in Computer Science & Engineering dedicated to the broad domain of AI for Social Good[cite: 7]. My research bridges two remarkably diverse worlds: the generative realm of creative writing and Natural Language Processing (NLP)[cite: 7], and the rigorous, physics-bound domains of Signal Processing and Quantitative Biostatistics.
Whether I am engineering physics-grounded deepfake forensics to protect digital ecosystems or analyzing complex auditory signals and biomedical data, my goal remains the same: building transparent, verifiable technologies that serve public health, combat digital deception, and empower human-computer interaction[cite: 7].
Future Directions
- AI for Social Good: Unifying rigorous computational methods into verifiable, human-centered tools that explicitly protect digital ecosystems and empower everyday users[cite: 7].
- Quantitative Biostatistics: Applying robust non-parametric statistical fusion (like MAD) to model physiological and auditory signals, developing reliable AI diagnostics[cite: 6].
- Signal & Audio Processing: Investigating complex acoustic environments and time-frequency domains (FFT) to detect biological anomalies, voice emotions, and synthetic speech[cite: 6, 7].
- Media Forensics & Deepfakes: Advancing multi-modal threat detection with a specific focus on combating auditory/visual manipulation in social networks[cite: 6, 7].
Research Skills
Signal & Audio Processing
NLP & Language Models
Biostatistics & Forensics
Selected Research
PHORENSICS: Physics-Inspired Forensics
Sub-Domain: Signal Processing & Quantitative Biostatistics
A deterministic, multi-modal unbiased forensics framework that replaces probabilistic black-box ML models. It leverages extensive time-domain signal analysis and frequency-domain statistics to establish strict mathematical boundaries against deepfakes.[cite: 7, 6]
Adaptive AI Fact Verification
Sub-Domain: NLP & Human-AI Alignment
Developed a retrieval-augmented fact verification framework combining Gemma 3 LLM reasoning with semantic document retrieval[cite: 7]. Proposed an adaptive convergence strategy that terminates retrieval once semantic similarity stabilizes, protecting against cognitive overload.
BEN-RS-ANN: Bengali Emotion & Speech
Sub-Domain: Auditory Signals & NLP
Developed a resource-scarce Artificial Neural Network (RS-ANN) designed for emotion detection in Bengali text and synthesized speech[cite: 7]. This framework uniquely aligns creative writing sentiment with auditory cues to resolve complex polysemy.
Research Experience
Research Intern (Media Forensics)
University of Calcutta | Advised by Prof. Sanjit Kumar Setua
- Worked on verifiable media forensics utilizing signal processing to protect users from manipulated digital media and synthetic content[cite: 7].
- Designed a fault-tolerant execution pipeline dynamically switching between GPU and CPU to ensure system accessibility across varying hardware capabilities[cite: 7].
Research Intern (Academic)
University of Calcutta | Advised by Prof. Rajat Kumar Pal
- Developed a Bengali NLP project for emotion detection, focusing on polysemy resolution in resource-scarce environments to improve human-language understanding[cite: 7].
- Evaluated using 10-fold cross-validation, achieving 90.08% accuracy[cite: 7].
Research Collaborator
Indian Statistical Institute (ISI) & The Bhawanipur Education Society College
- Collaborated on a machine-learning recommendation system built to respect user localization parameters without aggressively mining user-history data[cite: 7].
Invited Talks
Peer-Reviewed Publications
BEN-RS-ANN: An Innovative Approach for Revealing Emotion from Bengali Text[cite: 7]
In Applied Computing for Software and Smart Systems (ACSS), Springer LNNS, 2024[cite: 7]
Co-authored the proposal of an RS-ANN architecture contributing directly to mathematical polysemy resolution[cite: 7].
A User Independent Recommendation System for Web Series[cite: 7]
In Emerging Technologies in Data Mining and Information Security (IEMIS), Springer LNNS, 2022[cite: 7]
Co-developed a personalized recommendation engine mitigating external dataset bias[cite: 7].