Summer Research Internship 2026
DAU announces Summer Research Internship for non-DAU students from May to July 2026.
Student Eligibility
- Currently enrolled UG, PG and PhD students in good academic standing at accredited national institutions are eligible to apply (students currently enrolled at DAU are not eligible for this program).
- Specific eligibility criteria, such as minimum CPI, completion of specific coursework and prerequisites, the degree program, etc., will be determined by the individual DAU faculty mentor.
Internship Structure and Duration
- The SRI Program will typically run for a period of eight weeks between May and July. Specific dates each year will be determined by a faculty mentor.
- Interns are expected to commit to a minimum of 40 hours per week to their research project, as agreed upon with their faculty mentor.
- The research project will typically require the student to engage in various aspects of the research process, such as literature review, data collection, data analysis, and potentially manuscript preparation or presentation.
Stipend and Housing
- The intern shall receive a stipend of Rs. 1000 per week for a maximum of eight weeks.
- The intern shall reside on the DAU campus. The hostel accommodation charges shall be waived.
- The Institute will provide a travel allowance of up to INR 3000. The selected outstation candidate will need to submit the original receipts/tickets for travel to/from DAU during the internship period.
- Meals and other expenses shall be the responsibility of the intern.
Application Process
- Interested students shall directly submit a formal application to the DAU faculty mentor through email.
- The application shall include:
- The student’s resume or curriculum vitae highlighting their academic achievements and relevant experiences.
- A statement of interest outlining their research interests, relevant coursework, and reasons for applying to the specific internship(s).
- No objection certificate from their parent Institute.
- Optionally, letters of recommendation.
Download Flyer with QR Code for SRI 2026
Summer Research Internship Projects 2026 @ DAU
| Sr. No | Faculty Name | Research Area | Project Title | Project Abstract | Desired Skills | Position Status | |
|---|---|---|---|---|---|---|---|
| 1 | Prof. Abhishek Jindal | Applied Machine Learning, Wireless Communication, Finance and Cyber Security | Data management and analytics for fintech applications | The work involves predictive analytics, platform development, data collection and curation, and a deployment plan for an interesting problem in the fintech domain. | Data analytics using Python, Web development, Web scrapping and Project proposal drafting | abhishek_jindal[at]dau[dot]ac[dot]in | Open |
| 2 | Prof. Abhishek Jindal | Applied Machine Learning, Wireless Communication, Finance and Cyber Security | Application of Deep learning and NLP to Finance | The work involves developing new techniques of portfolio optimisation and stock/bitcoin trading, including futures and options. Project outcome may include a primitive platform development. | Basic knowledge of trading, futures and options, Deep learning, applied NLP | abhishek_jindal[at]dau[dot]ac[dot]in | Open |
| 3 | Prof. Abhishek Jindal | Applied Machine Learning, Wireless Communication, Finance and Cyber Security | Cybersecurity for enterprise systems | The work involves the identification of the vulnerabilities of an enterprise system and developing solutions for its mitigation. AI-based approaches and recent studies with available datasets will be the focus of the study. | Techniques for risk identification and mitigation, Deep learning and product development. | abhishek_jindal[at]dau[dot]ac[dot]in | Open |
| 4 | Prof. Ajay Beniwal | Printed and Flexible Electronics/Smart Sensing Technologies | Development and Characterization of a Temperature Sensor | Develop and characterise temperature sensors, gaining hands-on experience in fabrication and measurement. | Prerequisite Skills: Basic knowledge of electronics and circuits, familiarity with sensors, and interest in hands-on experimental work. | ajay_beniwal[at]dau[dot]ac[dot]in | Open |
| 5 | Prof. Ajay Beniwal | Printed and Flexible Electronics/Smart Sensing Technologies | IoT/Arduino-Integrated Smart Sensing Platform | The project offers hands-on experience in smart sensing devices, electronics integration, and data collection. Sensing device will be integrated with Arduino or IoT systems to enable real-time experimental monitoring. | Prerequisite Skills: Basic electronics and circuit knowledge, Arduino/microcontroller familiarity, and interest in hands-on experimental work. | ajay_beniwal[at]dau[dot]ac[dot]in | Open |
| 6 | Prof. Ankit Vijayvargiya | Biomedical Signals, Machine Learning, Neural Rehabilitation, Gait Analysis | AI-Driven ECG Image Digitization and Signal Processing-Based Arrhythmia Classification System | This project develops an AI-driven system for digitizing ECG images and detecting arrhythmias using signal processing. Scanned ECG images are converted into digital signals through preprocessing and waveform extraction. The signals are enhanced using filtering and R-peak detection, and features are classified using machine learning models. The system aims to automate ECG interpretation and support clinical decision-making. | Python Programming, Machine Learning, Image Processing | ankit_vijayvargiya[at]dau[dot]ac[dot]in | Open |
| 7 | Prof. Ankit Vijayvargiya | Biomedical Signals, Machine Learning, Neural Rehabilitation, Gait Analysis | Hierarchical Hand Gesture Recognition using Detection and Classification | This project proposes a two-stage hierarchical hand gesture recognition system based on signal processing and classification. This hierarchical approach reduces classification complexity and improves recognition accuracy by narrowing the search space. The proposed method enables efficient gesture understanding for human–computer interaction, assistive technologies, and intelligent robotic control applications. | Python Programming, Machine Learning | ankit_vijayvargiya[at]dau[dot]ac[dot]in | Open |
| 8 | Prof. Ankit Vijayvargiya | Biomedical Signals, Machine Learning, Neural Rehabilitation, Gait Analysis | AI-Powered Early Detection of Musculoskeletal Disorders in Garment Industry Operators | This project proposes an AI-powered system for early detection of musculoskeletal disorders in garment industry operators using posture analysis. Skeletal landmarks and joint angles are extracted to analyze repetitive movements and sustained postures. Machine learning models assess ergonomic risk levels and provide early warnings. The system aims to improve worker safety, reduce productivity loss, and support proactive occupational health monitoring. | Python Programming, Machine Learning, Basic computer vision concepts | ankit_vijayvargiya[at]dau[dot]ac[dot]in | Open |
| 9 | Prof. Arunava Chakravarty | Medical Image Analysis with Deep Learning | Exploring Latent Flow Models for Anatomical Structure Segmentation | Medical image segmentation plays a crucial role in clinical applications such as disease diagnosis, treatment planning, and monitoring anatomical changes over time. Accurately segmenting multiple structures is challenging due to variability in shape, appearance, and inter-relationships among structures. In this project, the student will explore a novel approach that leverages latent space representations of anatomical shapes. Using implicit neural representations (INRs), anatomical structures will be encoded in a compact latent space, and a flow-based model will be trained to transform these latent shapes according to the input image. This project will allow the student to investigate whether segmenting structures in latent space can improve accuracy, smoothness, and generalization, compared to conventional pixel-space segmentation. Experiments will be conducted on publicly available medical image datasets across modalities such as MRI, CT, or retinal OCT, focusing on the segmentation of multiple anatomically related structures. The project offers experience in modern deep learning approaches, generative modeling concepts, and shape-aware medical image analysis. |
Basic programming skills in Python, familiarity with deep learning concepts (Convolutional Neural Networks, Vision Transformers) and the ability to implement models in PyTorch. An interest in generative modeling and shape-aware methods is essential. Experience with medical imaging modalities (MRI, CT, OCT, etc.) is a plus but not required. | arunava_chakravarty[at]dau[dot]ac[dot]in | Open |
| 10 | Prof. Arunava Chakravarty | Medical Image Analysis with Deep Learning | Continuous-Time Modeling of Deformable Level Set Models for Multi-Structure Segmentation | Deformable contour models, such as level sets, provide a natural framework for anatomical segmentation by leveraging shape priors and inter-structure relationships. Recent deep learning approaches, such as recurrent active contour networks (RACEnet), modeled these contours using discrete-time RNNs, but training efficiency and stability remain challenging. In this project, the student will explore a continuous-time formulation using Neural Ordinary Differential Equations (Neural ODEs) to model the evolution of deformable contours, enabling stable and efficient segmentation of multiple interacting structures. The project will focus on developing methods to incorporate shape priors and coupled contour representations for multi-object segmentation, while maintaining computational efficiency. Experiments will be performed on publicly available medical image datasets, such as cardiac MRI, abdominal CT, or retinal OCT, giving the student hands-on experience with continuous-time modeling, differentiable numerical methods, and anatomically informed segmentation. |
Basic programming skills in Python, familiarity with deep learning concepts (Convolutional Neural Networks, Vision Transformers) and the ability to implement models in PyTorch. An interest in generative modeling and shape-aware methods is essential. Experience with medical imaging modalities (MRI, CT, OCT, etc.) is a plus but not required. | arunava_chakravarty[at]dau[dot]ac[dot]in | Open |
| 11 | Prof. Hemant A. Patil | Speech and Audio Research | Audio Deepfake Detection and Speaker Identification | Audio deepfake detection (ADD) when combined with speaker identification system is also known as Spoofing-aware Automatic Speaker Verification (SASV), which focuses on a very practical and increasingly important problem: modern speaker verification systems can be easily fooled by synthetic or manipulated speech also known as deepfake audio which are generated by various GenAI methods such as voice cloning, Text-To-Speech (TTS) or Generative Adversarial Networks (GANs). In a real-world setting, it’s not enough to just verify who is speaking; the system must also decide whether the speech itself is genuine or spoofed. This turns the task into a joint problem where speaker verification and spoof detection have to work together, not independently. In this project, we aim to build systems that remain reliable even when faced with unseen attack types or changing recording conditions. Rather than optimizing isolated components, we focus on the full SASV pipeline and how its parts interact under real-world constraints. Concretely, the work involves: Developing robust countermeasure (CM) models, building strong ASV systems, exploring feature representations and End-to-End (E2E) systems, designing fusion strategies for SASV, improving generalization and robustness of models, developing and analyzing evaluation metrics for SASV systems, error analysis and explainability of deepfake detection models, benchmarking and reproducibility, towards deployable systems trained using federated learning, performing segment-wise analysis on partially spoofed (Half-Truth) audio. Publications with previous interns: [1] Shah, A.J., Pandey, A., Gaikwad, M.A. and Patil, H.A., 2025, October. A Wavelet Tour of Audio Deepfake Detection. In 2025 Asia Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC) (pp. 2247-2252), 2025. [2] Shah, Arth J., Aniket Pandey, Satyam R. Tiwari, and Hemant A. Patil. "An Enhanced Probabilistic Approach for Singfake Generation." In 2025 Asia Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC), pp. 2086-2091, 2025. [3] Mahyavanshi, R., Reddy, C.M., Shah, A.J. and Patil, H.A., 2024, December. Teager energy cepstral coefficients for audio deepfake detection. In 2024 Asia Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC) (pp. 1-6). |
Machine Learning, Signal Processing, Deep Learning (optional), however hunger to learn and curiosity to explore more is must !!! ;) - Prof. Patil | hemant_patil[at]dau[dot]ac[dot]in | Open |
| 12 | Prof. Hemant A. Patil | Dysarthric Severity-Level Classification | Dysarthria is a motor speech disorder caused by neurological impairments, leading to reduced speech intelligibility and variability in severity across individuals. This project aims to develop an automated dysarthric severity-level classification system using advanced signal processing and Machine Learning/Deep Learning techniques. The system will extract robust acoustic features such as Mel-Frequency Cepstral Coefficients (MFCCs), pitch, formants, jitter, shimmer, and spectral features from speech signals. These features will be used to train and evaluate classification models including traditional ML algorithms (e.g., Support Vector Machines, Random Forests) and Deep Learning architectures such as Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and Transformer-based models. The project may also explore end-to-end learning from raw waveforms and incorporate data augmentation techniques to handle limited clinical datasets. Performance will be evaluated using standard metrics such as accuracy, F1-score, and confusion matrices. The final system aims to assist clinicians in objective and scalable assessment of dysarthria severity, enabling early diagnosis, monitoring, and personalized therapy planning. Publications with previous interns: [1] Avula, M., Pusuluri, A. and Patil, H.A., 2024, December. Significance of entropy based features for dysarthric severity level classification. In 2024 Asia Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC) (pp. 1-6). [2] Mannepalli, R.S., Pusuluri, A. and Patil, H.A., 2024, December. Dysarthria Severity Classification Using Phase Based Features of LP Residual. In 2024 Asia Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC) (pp. 1-5). [3] Sahasra, G. S., Swapna, K., Srivastava, A., Pusuluri, A., & Patil, H. A. (2024, December). Comparative Analysis of Glottal and Vocal Tract Features in Dysarthria. In 2024 Asia Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC) (pp. 1-6). |
Students are expected to have basic programming knowledge (preferably in Python) and a willingness to learn new tools and concepts during the project. Familiarity with fundamental concepts in mathematics (such as linear algebra and probability) and basic signal processing or data analysis will be helpful but not strictly required. | hemant_patil[at]dau[dot]ac[dot]in | Open | |
| 13 | Prof. Hemant A. Patil |
Speech and Audio Research |
Audio Large Language Models | With the growing need for assistive speech technology, it is imperative to develop Audio LLMs that can better understand dysarthric speech and support communication for people with speech impairments. Current assistive speech recognition systems often fail to accurately transcribe severely distorted speech, leaving dysarthric individuals frustrated and excluded from digital communication tools. This project addresses this gap by reviewing recent advances in audio-visual speech processing and developing a working model that reconstructs the intended vocal plan from distorted audio-video input, then uses a plan-conditioned language model to predict the final text with higher accuracy. The problem: Standard speech recognition works well for healthy speech but struggles with dysarthria because it treats distortion as random noise. People with dysarthria know what they want to say, but their motor control makes it hard to produce clear speech. The result is poor recognition rates and limited accessibility. Our solution The model works in two steps: Reverse‑engineer the vocal plan from audio features and mouth movement landmarks. Predict text using a language model conditioned on that vocal plan. This makes the system more robust and interpretable, helping both recognition and clinical understanding. Why it matters This project creates a more inclusive speech technology that can: improve daily communication for dysarthric users, support speech therapy with interpretable feedback, enable better assistive devices and services. Publications with previous interns: [1] Trivedi, N., Purohit, R.M., Tiwari, S. and Patil, H.A., 2025, August. Cosshi: Llm-powered large-scale multitask hinglish dataset for speech forensics. In 2025 International Conference on Asian Language Processing (IALP) (pp. 135-140). |
Applicants should have Basic programming skills (preferably in Python) and an interest in AI, speech, or multimodal systems. Familiarity with fundamental Machine Learning or Deep Learning concepts is helpful but not mandatory. Exposure to libraries such as NumPy, PyTorch/TensorFlow, or basic audio processing (e.g., spectrograms handling) will be an advantage. Students should be willing to explore concepts like speech recognition, Audio LLMs, and multimodal learning (audio + text), and work with real-world datasets. | hemant_patil[at]dau[dot]ac[dot]in | Open |
| 14 | Prof. Hemant A. Patil |
Speech and Audio Research |
Text-To-Speech (TTS)/ Voice Conversion/ Voice or Speech Agents for next-gen Intelligent Systems | In this module, the motivation is to adapt the western TTS models for the Indic languages in order to create the expressive TTS in Indian language. It includes the building, deploying and optimising of the TTS from scratch and at last, we plan to proposed the novel architecture in order to synthesise the natural utterances, which able to controls/preserves the basic properties of the speech, e.g., Prosody, Emotions, Flow. Furthermore, based on output, we will try to integrate with agent, to improve applicability in domain. Deliveries: 1. Open source TTS for global south community. 2. Intellectual Property rights. 3. Possible submission to prestigious journals or conferences. Publications with previous interns: [1] Gajre, K., Gupta, R., Purohit, R.M. and Patil, H.A., 2025, October. NAMTalk: From Muscle Vibrations to Emotional Speech. In International Conference on Speech and Computer (pp. 113-128). Cham: Springer Nature Switzerland. |
Applicants should have Basic programming experience (preferably in Python) and an interest in speech processing, AI, or Deep Learning. Familiarity with fundamental concepts in Machine Learning, neural networks, or audio processing is helpful but not mandatory. Exposure to tools such as NumPy, PyTorch/TensorFlow, or basic signal processing concepts (like spectrograms or frequency analysis) will be an advantage. Students should be willing to learn about Text-to-Speech (TTS), experiment with models, and work with real-world datasets. Curiosity, consistency, and the ability to explore research papers and implement ideas are more important than prior expertise. | hemant_patil[at]dau[dot]ac[dot]in | Open |
| 15 | Prof. Hemant A. Patil |
Speech and Audio Research |
Source Tracing for Deepfake Audio | Deepfake audio has become increasingly realistic, making it difficult not only to detect whether a clip is fake, but also to understand where it came from. Source tracing of deepfake audio focuses on this next step, identifying the underlying generative model, features used, or process used to create the deepfake audio. This is important for attribution, forensic analysis, and building accountability in real-world scenarios where multiple synthesis methods (e.g., TTS, voice conversion, GANs diffusion-based models) are constantly evolving. The challenge lies in the fact that different models can produce very similar outputs, and artifacts that indicate the source are often subtle, inconsistent, or disappear under compression and noise. In this project, we aim to go beyond binary detection and build systems that can reliably trace the origin of deepfake audio, even under unseen conditions and across diverse generation methods. The work includes: Build robust source tracing (or) source verification architectures. Learn discriminative representations to capture spoofing artifacts for source attribution. Dataset creation and curation for source tracing. Source attribution under degradation and noisy conditions simulating real-life environments. Open-set source tracing. Evaluation metrics and protocols. Analysis and interpretability. Towards forensic applications. Publications with previous interns (entire list of 12 publications on website, below are few listed): [1] Shah, Arth J., Aniket Pandey, Satyam R. Tiwari, and Hemant A. Patil. "An Enhanced Probabilistic Approach for Singfake Generation." In 2025 Asia Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC), pp. 2086-2091, 2025. [2] Shah, A.J., Pandey, A., Gaikwad, M.A. and Patil, H.A., 2025, October. A Wavelet Tour of Audio Deepfake Detection. In 2025 Asia Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC) (pp. 2247-2252), 2025. [3] Gajre, K., Gupta, R., Purohit, R.M. and Patil, H.A., 2025, October. NAMTalk: From Muscle Vibrations to Emotional Speech. In International Conference on Speech and Computer (pp. 113-128). Cham: Springer Nature Switzerland. [4] Trivedi, N., Purohit, R.M., Tiwari, S. and Patil, H.A., 2025, August. Cosshi: Llm-powered large-scale multitask hinglish dataset for speech forensics. In 2025 International Conference on Asian Language Processing (IALP) (pp. 135-140). [5] Mahyavanshi, R., Reddy, C.M., Shah, A.J. and Patil, H.A., 2024, December. Teager energy cepstral coefficients for audio deepfake detection. In 2024 Asia Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC) (pp. 1-6). [6] Sahasra, G. S., Swapna, K., Srivastava, A., Pusuluri, A., & Patil, H. A. (2024, December). Comparative Analysis of Glottal and Vocal Tract Features in Dysarthria. In 2024 Asia Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC) (pp. 1-6). [7] Avula, M., Pusuluri, A. and Patil, H.A., 2024, December. Significance of entropy based features for dysarthric severity level classification. In 2024 Asia Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC) (pp. 1-6). [8] Mannepalli, R.S., Pusuluri, A. and Patil, H.A., 2024, December. Dysarthria Severity Classification Using Phase Based Features of LP Residual. In 2024 Asia Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC) (pp. 1-5). [9] Shah, A.J., Yadav, S.H. and Patil, H.A., 2024, December. Teager Energy Cepstral Coefficients for Spoken Language Identification. In 2024 Asia Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC) (pp. 1-6). [10] Shah, Arth J., Manish Suthar, and Hemant A. Patil. "Multi-Block U-Net for Wind Noise Reduction in Hearing Aids." In International Conference on Pattern Recognition, pp. 234-249. Cham: Springer Nature Switzerland, 2024. |
Machine Learning, Signal Processing, Deep Learning (optional), however hunger to learn and curiosity to explore more is must !!! ;) - Prof. Patil | hemant_patil[at]dau[dot]ac[dot]in | Open |
| 16 | Prof. Minal Bhise |
Databases, Distributed Databases : Query Optimization, Resource Estimation, Data Analytics |
Lightweight Resource Estimation for Modern Databases | Modern data-driven applications generate large volumes of data that must be processed efficiently to extract useful insights. Query processing over large datasets requires significant computational resources such as CPU, memory, and I/O bandwidth. Traditional database systems rely on cost-based query optimizers that often fail to accurately estimate resource requirements for dynamic workloads. This project plans to apply appropriate machine learning techniques for query optimization and resource estimation for modern databases. | Databases, Basics of Machine Learning, Programming | minal_bhise[at]dau[dot]ac[dot]in | Open |
| 17 | Prof. Minal Bhise |
Databases, Distributed Databases: Query Optimization, Resource Estimation, Data Analytics |
Dynamic Query Processing | Handling queries efficiently at the edge is challenging due to changing workloads. This project aims to predict future query load using machine learning approaches so that the system can plan ahead, reduce delays, and utilize resources efficiently. | Databases, Basics of Machine Learning, Programming | minal_bhise[at]dau[dot]ac[dot]in | Open |
| 18 | Prof. Minal Bhise |
Databases, Distributed Databases: Query Optimization, Resource Estimation, Data Analytics |
Edge Intelligence | Edge devices have limited power and memory, so not all tasks can be processed locally. This project decides when to execute tasks at the edge and when to offload them to improve performance and reduce system load using appropriate machine learning methods. | DBMS, Programming, Data Structures, Basics of Machine Learning | minal_bhise[at]dau[dot]ac[dot]in | Open |
| 19 | Prof. Pushpendra Kumar |
Neural networks, Mathematical modelling, Differential Equations |
Stability Analysis and Numerical Implementation of Fractional Order Neural Networks | The aim of this project is to analyze the stability of fractional-order neural networks using Lyapunov methods, comparison principles, and Mittag-Leffler-type stability concepts. It further seeks to utilize established results from fractional-order differential equations to derive sufficient conditions ensuring the stability of such neural network models. Additionally, the project aims to bridge theoretical analysis with practical implementation by integrating mathematical findings with hardware realization. This approach will help demonstrate the applicability and effectiveness of fractional-order neural network models in real-world scenarios. | Basics of neural networks, MATLAB | pushpendra_kumar[at]dau[dot]ac[dot]in | Open |
| 20 | Prof. Pushpendra Kumar |
Neural networks, Mathematical modelling, Differential Equations |
Analysis and Control of Infectious Diseases Using Fractional-Order Epidemic Models | This project focuses on the development and analysis of fractional-order epidemic models to better capture memory and hereditary effects in disease dynamics. Unlike classical integer-order models, fractional derivatives provide a more realistic description of infection spread over time. The study aims to investigate stability, existence, and control strategies for such systems. Numerical methods will be employed to simulate real-world scenarios and validate the effectiveness of the proposed models. The outcomes can help in designing improved intervention policies for epidemic control. | Basics of mathematical modelling, MATLAB | pushpendra_kumar[at]dau[dot]ac[dot]in | Open |
| 21 | Prof. Srimanta Mandal |
Computer Vision |
Computer Vision Based Physical Rehabilitation Assessment | Computer vision can help assess physical rehabilitation in a simple and accurate way. Using a camera, the system tracks a patient’s body movements during exercises. It compares these movements with correct patterns and gives feedback on posture, speed, and accuracy. This helps doctors and therapists understand the patient’s progress without always being physically present. It also allows patients to perform exercises at home with guidance. Such systems are low-cost, non-invasive, and easy to use. Overall, computer vision makes rehabilitation more accessible, efficient, and personalized, helping patients recover faster and perform exercises correctly with continuous monitoring and feedback. | Python Programming, Basics in Linear Algebra, Probability | srimanta_mandal[at]dau[dot]ac[dot]in | Open |
| 22 | Prof. Srimanta Mandal | Computer Vision | Computer Vision Applications in Bio-Medical Domain | Computer vision plays an important role in the biomedical domain by helping computers understand medical images and videos. It is widely used in tasks such as disease detection, medical image analysis, and patient monitoring. For example, it can analyze X-rays, MRI, and CT scans to detect abnormalities like tumors or fractures. It also helps in tracking body movements for rehabilitation and monitoring patients in hospitals. These systems reduce human effort, improve accuracy, and support doctors in making better decisions. Overall, computer vision makes healthcare more efficient, accessible, and reliable, leading to faster diagnosis, improved treatment, and better patient outcomes. | Python Programming, Basics in Linear Algebra, Probability | srimanta_mandal[at]dau[dot]ac[dot]in | Open |
| 23 | Prof. Srimanta Mandal | Computer Vision | Multi-modal Adversarial Attack and Defense | Multi-modal systems that use both audio and visual data are widely used in applications such as speech recognition, video analysis, and human–computer interaction. However, these systems are vulnerable to adversarial attacks, where small, carefully designed changes in audio or visual inputs can mislead the model into making wrong predictions. This work studies adversarial attacks in multi-modal settings and explores how disturbances in one or both modalities can affect system performance. It also discusses defense strategies, such as robust training and cross-modal verification, to improve security. Overall, the goal is to build reliable multimodal systems that remain accurate even under adversarial conditions. | Python Programming, Basics in Linear Algebra, Probability | srimanta_mandal[at]dau[dot]ac[dot]in | Open |
| 24 | Prof. Yash Vasavada | Communications, Signal Processing and Machine Learning | Blind channel estimation and symbol detection | A data-aided scheme is proposed where the unknown informative symbols and channel coefficients are estimated in an iterative manner. | Signal processing, linear algebra and communication | yash_vasavada[at]dau[dot]ac[dot]in | Open |
| 25 | Prof. Yash Vasavada | Communications, Signal Processing and Machine Learning | A sparsely coded power domain non-orthogonal multiple access | A novel NOMA method is proposed based on a combination of code-domain and power-domain non-orthgonal multiplexing of user symbols. | Wireless communication, signal processing | yash_vasavada[at]dau[dot]ac[dot]in | Open |
| 26 | Prof. Yash Vasavada | Communications, Signal Processing and Machine Learning | Sparse matrix precoded MIMO | A novel MIMO scheme is proposed based on sparse matrix precoding. | Communications and signal processing | yash_vasavada[at]dau[dot]ac[dot]in | Open |
| 27 | Prof. Yash Vasavada | Communications, Signal Processing and Machine Learning | A machine learning scheme for anomaly detection in electricity consumption data | Development of AI-based method for detecting outliers and anomalies in electricity load time series. | AI and ML | yash_vasavada[at]dau[dot]ac[dot]in | Open |
| 28 | Prof. Manish Chaturvedi | Applications of IoT and ML in designing Intelligent Transportation System solutions | Last Mile connectivity for Public Transportation System | This work will explore modeling and simulation based analysis of various last mile connectivity solutions, including shared rides, and active modes like bicycles and walking. Please find more information about the work at Manish Chaturvedi and Sanjay Srivastava, "A Multi-modal Ride Sharing Framework for Last Mile Connectivity," 14th International Conference on COMmunication Systems \& NETworkS (COMSNETS), IEEE/ACM, 2022, pp. 824-829, Venue: Online, DOI: 10.1109/COMSNETS53615.2022.9668583. | Problem solving skills, Programming, Algorithm design | manish_chaturvedi[at]dau[dot]ac[dot]in | Open |
| 29 | Prof. Manish Chaturvedi | Applications of IoT and ML in designing Intelligent Transportation System solutions | In-door localization and routing solutions | This work explores solutions for Indoor localization using opportunistically available information from various sources and selectively deployed infrastructure elements in large and diverse indoor environments. | Good Problem solving skills, Programming, Basic Image processing | manish_chaturvedi[at]dau[dot]ac[dot]in | Open |
| 30 | Prof. Gopinath Panda | Queueing Games, Bayesian estimation | Stability Analysis Queues via Bayesian Credible Intervals | In classical queueing theory, a system is considered stable only if the traffic intensity is smaller than 1. However, point estimates from Maximum Likelihood Estimation (MLE) often fail to account for the risk of "momentary instability" when data is scarce. This project focuses on the theoretical derivation of the posterior distribution for using conjugate Gamma priors. The student will investigate how different "prior beliefs" (from highly certain to "non-informative") influence the probability that a system is classified as stable. By calculating the Posterior Risk under different loss functions, the study provides a mathematical framework for determining system reliability when arrival and service rates are uncertain. |
Basic Probability, Inferential Statistics, Stochastic Models | gopinath_panda[at]dau[dot]ac[dot]in | Open |
| 31 | Prof. Gopinath Panda | Queueing Games, Bayesian estimation | Bayesian Analysis of Peak-Hour Queueing Dynamics at the DAU Cafeteria | Cafeteria congestion is a daily challenge for university students, often resulting in abandoned queues (balking, reneging, jockeying) and dissatisfaction. This project utilises Bayesian estimation within a single-server M/M/1 queueing model to analyze and predict service efficiency during peak lunch hours. Unlike traditional methods that depend on large datasets, the Bayesian approach combines "expert knowledge" (prior insights from cafeteria staff) with real-time observational data to produce accurate wait-time distributions. We will further improve it to a multi-server queue for more realistic scenarios. By the end of the study, the student will quantify the probability of line overflows and provide data-driven recommendations for optimal staffing levels to reduce average wait times for students. | Probability, Bayesian Statistics, Stochastic Models | gopinath_panda[at]dau[dot]ac[dot]in | Open |
| 32 | Prof. Gopinath Panda | Queueing Games, Bayesian estimation | Blockchain Queueing Games | In public blockchains, users compete to have their transactions included in limited block space by bidding "gas" or transaction fees. This project models the Mempool as a priority queue where the arrival rate and service rate are influenced by strategic bidding. Using a Bayesian Game framework, the student will estimate how users update their "fee beliefs" based on observed congestion levels. The study aims to determine the Nash Equilibrium—the point where no user can reduce their wait time without overpaying—and analyze how "prior" knowledge of network spikes affects overall queue stability and miner revenue. | Bayesian Statistics, Blockchain, Python | gopinath_panda[at]dau[dot]ac[dot]in | Open |
| 33 | Prof. Gopinath Panda | Queueing Games, Bayesian estimation | Toll-Based Regulation of Selfish Routing in Congested AI Service Systems | In open-access queueing systems, such as public AI chatbots or cloud APIs, individual users make "selfish" decisions about when to join the queue based solely on their own interests. This behavior often results in a negative externality, causing the queue length to exceed the socially optimal level and diminishing the overall efficiency of the network—a situation reminiscent of the "Tragedy of the Commons" in a digital context. This project builds upon the foundational model established by Naor (The Regulation of Queue Size by Levying Tolls) by exploring the effects of Dynamic Tolls as a regulatory mechanism. Through mathematical derivation and simulation, the student will determine the optimal fee that aligns an individual's selfish interests with the socially optimal outcome. The study aims to demonstrate that a well-calibrated toll can maximize social welfare by discouraging low-value users from joining during peak congestion, thereby ensuring faster response times for critical tasks. |
Game theory, Queueing, Python | gopinath_panda[at]dau[dot]ac[dot]in | Open |
| 34 | Prof. Pankaj Kumar | RF and Microwave, VLSI | Solar Energy Harvesting Using Metasurface | Solar energy harvesting using metasurfaces has emerged as a promising approach to enhance light–matter interaction and improve energy conversion efficiency beyond conventional photovoltaic systems. In this study, a compact and efficient metasurface-based solar energy harvesting structure is proposed, designed to maximize broadband absorption across the visible and near-infrared spectrum. The metasurface consists of periodically arranged subwavelength resonators engineered to achieve near-unity absorption through impedance matching and resonant coupling mechanisms | Should have basic understanding of electromagnetics wave | pankaj_kumar[at]dau[dot]ac[dot]in | Open |
| 35 | Prof. Sudip Bera | Algebraic graph theory | On the domination number of graphs | The study of graphs associated with various algebraic structures has been a topic of active investigation during the last two decades. There are several benefits of studying such graphs as they help to realize the interdependence between the algebraic structures and the corresponding graphs. Our goal is to contribute to the study of the domination number of graphs arising from algebraic structures. | A background in combinatorics, graph theory, and group theory is required. Familiarity with Sage programming will be an added advantage. | sudip_bera[at]dau[dot]ac[dot]in | Open |
| 36 | Prof. Prosenjit Kundu | Nonlinear Dynamics, Complex Networks | Resilience in higher-order networks | In an increasingly interconnected world, the stability of complex systems depends not merely on their efficiency but on their resilience, their capacity to absorb disturbances, reorganise, and continue functioning under stress. Complex networks, spanning domains such as technological infrastructures, ecological systems, and social interactions, exhibit emergent behaviours shaped by intricate, non-linear relationships among their components. These characteristics make them both powerful and inherently vulnerable to cascading failures. This work situates resilience as a higher-order property that arises from the interplay between network structure, dynamics, and adaptive capacity. Rather than viewing resilience solely as resistance to failure, it reframes it as an evolving capability involving robustness, flexibility, and recovery. The discussion critically examines how topological features such as connectivity distribution, modularity, and redundancy govern the propagation of disruptions and the system’s ability to reconfigure itself. By bridging theoretical insights with practical implications, this study underscores the necessity of designing networks that are not only optimized for performance but also capable of transformation under uncertainty. Ultimately, it advances a deeper understanding of resilience as a foundational principle for sustaining complex systems in the face of growing systemic risks. | Linear Algebra, Dynamical systems (Basic), Programing | prosenjit_kundu[at]dau[dot]ac[dot]in | Open |
| 37 | Prof. Prosenjit Kundu | Nonlinear Dynamics, Complex Networks | Failure in distribution networks: a complex network approach | Failures in distribution networks can be effectively analyzed using complex network theory, where systems are represented as interconnected nodes and links. This approach highlights how network topology such as connectivity patterns and node/edge importance affects system robustness and vulnerability. The study examines the impact of random failures and targeted attacks, showing that disruptions to critical nodes can trigger cascading failures and widespread outages. Methods such as percolation theory and resilience metrics are used to assess network stability and failure propagation. The abstract also emphasizes strategies to improve resilience, including network redundancy, decentralized design, and adaptive reconfiguration. Overall, a complex network perspective provides valuable insights for enhancing the reliability and robustness of modern distribution systems. | Linear Algebra, Graph theory/complex networks, programing | prosenjit_kundu[at]dau[dot]ac[dot]in | Open |
| 38 | Prof. Mukesh Tiwari and Prof. Prosenjit Kundu (@ CSys Lab) | Machine Learning and Dynamical Systems | Dynamics-Informed Reservoir Computing | Dynamics-Informed Reservoir Computing (DIRC) is an emerging approach for modeling complex dynamical systems by incorporating system-specific structural and temporal information directly into reservoir design. Traditional Reservoir Computing methods, such as Echo State Networks, rely on randomly generated recurrent connections, which often limit interpretability and may not fully capture the underlying system dynamics. In contrast, DIRC leverages prior knowledge or data-driven insights to construct reservoirs that reflect the intrinsic behavior of the target system. This work explores the integration of dynamical features such as temporal correlations, state-space structures, and nonlinear dependencies into reservoir architectures using techniques like graph-based representations and topology-aware connectivity. By embedding meaningful dynamical patterns into the reservoir, the framework enhances the system’s ability to learn, predict, and generalize complex behaviors, including chaotic and multi-scale processes. The effectiveness of the approach is demonstrated across a variety of benchmark problems, showing improvements in prediction accuracy, stability, and interpretability compared to conventional random reservoirs. Dynamics-Informed Reservoir Computing thus provides a principled and efficient paradigm for advancing data-driven modeling of complex systems. |
Machine Learning, Linear Algebra | mukesh_tiwari[at]dau[dot]ac[dot]in and prosenjit_kundu[at]dau[dot]ac[dot]in | Open |
| 39 | Prof. Tapas Kumar Maiti | Electronics, Robotics, and Cybernetics | Intelligent MOSFET | Intelligent MOS Devices (Metal-Oxide-Semiconductor devices) refer to advanced semiconductor components that combine traditional MOS structures (like MOSFETs) with AI/ML capabilities. These devices go beyond simple switching/amplification and act as smart, adaptive electronic components. | Basic Electronics and Computer Science | tapas_kumar[at]dau[dot]ac[dot]in | Open |
| 40 | Prof. Tapas Kumar Maiti | Electronics, Robotics, and Cybernetics | Data Analytics in Semiconductor Manufacturing and Logistics | Data Analytics in Semiconductor Manufacturing and Logistics is the application of statistical methods, machine learning, and real-time data processing to optimize fabrication processes, improve yield, and streamline supply chains. This is a critical pillar of modern semiconductor Gigafabs and Terafabs (e.g., Intel, TSMC, Samsung Electronics, Tesla), where even nanometer-level variations can impact billions of dollars. |
Basic Electronics and Computer Science | tapas_kumar[at]dau[dot]ac[dot]in | Open |
| 41 | Prof. Tapas Kumar Maiti | Electronics, Robotics, and Cybernetics | Embedded AI/ML | Integration of artificial intelligence (AI) and machine learning (ML) algorithms directly into low-cost embedded systems—small-scale, resource-constrained devices such as microcontrollers, sensors, or edge devices. Instead of sending data to the cloud for processing, these systems can analyze data locally in real time. | Basic Electronics and Computer Science | tapas_kumar[at]dau[dot]ac[dot]in | Open |
| 42 | Prof. Tapas Kumar Maiti | Electronics, Robotics, and Cybernetics | Chip Design for AI and ML | Chip Design for AI and ML focuses on building specialized hardware that can efficiently execute machine learning and deep learning algorithms. Unlike general-purpose CPUs, these chips are optimized for massive parallel computation, low power consumption, and high data throughput. | Basic Electronics | tapas_kumar[at]dau[dot]ac[dot]in | Open |
| 43 | Prof. Tapas Kumar Maiti | Electronics, Robotics, and Cybernetics | Robotics and Automation Systems | Robotics and automation technologies create powerful, autonomous systems capable of operating efficiently and securely at the edge. Key tools include ROS, TensorFlow Lite, TinyML, and platforms like NVIDIA Jetson and ESP32. | Electronics and Robotics | tapas_kumar[at]dau[dot]ac[dot]in | Open |
| 44 | Prof. Vinay S Palaparthy | AI/ML | Impact of hazardous gases on coal mine workers | The objective of this study is to collect the environmental parameters that arises in the coal mines and study its impact on the coal mine workers | M.Sc/B.Tech/M.tech (CS/IT/ECE/Electronics) | vinay_shrinivas[at]dau[dot]ac[dot]in | Open |
| 45 | Prof. Vinay S Palaparthy | AI/ML | EDA for the Sensor based data | In this project objective is to understand/collect the sensor data for the coal mines with the help of the in-house developed IoT-enabled sensor system | M.Sc/B.Tech/M.Tech (CS/IT/ECE/Electronics) | vinay_shrinivas[at]dau[dot]ac[dot]in | Open |
| 46 | Prof. Vinay S Palaparthy | AI/ML | Development of IoT-enabled sensor system for the coal mines | Testing the reliability of the of the in-house developed IoT-enabled sensor system under the in-situ conditions | M.Sc/B.Tech/M.Tech (CS/IT/ECE/Electronics/Physics) | vinay_shrinivas[at]dau[dot]ac[dot]in | Open |
| 47 | Prof. Vinay S Palaparthy | AI/ML | Green synthesis of the Nanomaterials & surface characterization for Coal mine sensors | To develop the sensing layer for the sensor required for the coal mines using the green synthesis method | M.Sc/B.Tech/M.Tech (CS/IT/ECE/Electronics/Physics) | vinay_shrinivas[at]dau[dot]ac[dot]in | Open |
| 48 | Prof. Vinay S Palaparthy | AI/ML | Server design for the IoT enabled sensor system | The objective of this project is to design the server for the data collected from the in-house developed sensor system. | M.Sc/B.Tech/M.Tech (CS/IT/ECE/Electronics/Physics) | vinay_shrinivas[at]dau[dot]ac[dot]in | Open |