AIMS-DTU

Artificial Intelligence and Machine Learning Society, DTU

Here, we focus on

Transferring Knowledge

We are a community of innovators, researchers, and developers dedicated to pushing the boundaries of AI and ML.

Instagram LinkedIn Discord

Knowledge Transfer

We believe in sharing knowledge, encouraging learning together as a team.

Revolutionizing Research

We turn ideas into meaningful research to make a difference.

Conquering Hackathons

It is a legacy of aims to lead Hackathons across the nation with innovative solutions.

OUR FLAGSHIPS

brAInwave

brAInwave is a 30-hour hackathon by AIMS-DTU, designed to push the limits of innovation.

Over two days, participants engage in insightful speeches, intense coding, and thorough project evaluations,expert mentorship, providing an ideal platform to showcase their skills and creativity.

With a reach of over 20K individuals, AIMS-DTU offers excellent visibility.

VisionX AI

Deep learning-based hackathon, challenging participants to develop groundbreaking solutions for real-world problems .

The primary objective is to foster a culture of ingenuity, collaboration, and continuous growth in the AI community. It consists of multiple tracks including NLP, Computer Vision and much more.

Synaptix

A high-stakes GenAI-focused competition with innovative problem statements, designed to push your technical creativity.

From ideation to live implementation, compete across two intense rounds judged by industry experts and professors.

IJCNN 2025

IllumiCurveNet: Low-Light Image Enhancement of Lunar Permanently Shadowed Regions Using a Self-Guided Loss Framework

Authors:

Saksham Jain • Sparsh Jain • Ashish Prajapati • Garvit Singh • Dinesh K. Vishwakarma

Overview:

IllumiCurveNet enhances lunar PSR images suffering from low visibility, poor contrast, and noise using an encoder–decoder with spatial attention, dilated convolutions, and adaptive gamma correction. A self-guided loss framework ensures texture preservation, contrast enhancement, and consistency. It achieves state-of-the-art results without paired data, aiding lunar mapping, navigation, and resource exploration.

Autonomous Cheating Detection for Online Examinations using Knowledge Distillation and Multitask Learning

Authors:

Vansh Sachdeva • Shashvat Singhal • Dinesh K. Vishwakarma

Overview:

This study proposes a deep learning-based real-time cheating detection system for online exams, integrating pose detection (PoseNet), eye tracking (Mediapipe Iris), and facial expression recognition (Vision Transformer). Using multi-task learning with knowledge distillation, the lightweight model achieved 83.5% accuracy and 0.82 F1-score, prioritizing recall to ensure secure remote assessments.

Audio Based Machine Fault Diagnosis using Hybrid Feature Extraction and Ensemble Learing

Authors:

Shashvat Singhal • Bhavya Goel • Kshitij Agrawal • Rithwick Sethi • Shashi Sah • Rachit Jain • Dinesh K. Vishwakarma

Overview:

This research detects machine faults through audio analysis, applying STFT and extracting features like Mel spectrogram, spectral kurtosis, and spectral centroid. These features are classified using algorithms including XBoost, SVC, and Random Forest. Testing on the MIMI dataset showed XBoost achieved the highest accuracy, reaching 98% in fault classification.

DSeP-xNet: A Feature Optimized Ensemble Framework for Hyperspectral-Based Soil Organic Carbon Prediction

Authors:

Akshyat Shah • Shashi Sah • Shashvat Singhal • Shagun Jain

Overview:

This paper presents DSeP-xNet, a soil organic carbon prediction framework combining preprocessing (Savitzky–Golay filtering, XGBoost feature selection) with a stacking ensemble of DNN, SE-CNN, and other models. By emphasizing noise removal and optimal feature selection, it enhances prediction accuracy, supporting smart agriculture through precise soil nutrient assessment.