• How AI can be  applied  in Healthcare 
• Exploring Real-World Use Cases of AI in Healthcare 
• Case Studies on : AI based diagnostics for : Cervical cancer, Gall bladder Cancer, Cardiac health , Breast Cancer, Mental fatigue detection, Neural Disorders  etc.

• How AI can be  applied  in Core Challenges in Business Analytics
• Exploring Real-World Use Cases of AI in Business Decision-Making
• Hands-On Training: Building and Applying AI Models for Business Insights
• Interactive Problem Formulation: Identifying and Defining AI-Driven Business Solutions

1. How AI Can Address Key Challenges in Social Sciences
2. Real-World Applications of AI in Social Science Domains
3. Interactive Session: Demonstrating AI Tools for Social Science
4. Leveraging AI for Social Good and Community Impact

1. Introduction to AI in Defence Technology.
2. How AI  can solve important problems in Defence Technology. 
3. Use Cases of AI and Defence Technology Working Together
4. Hands-On Demonstration: AI in Defence Technology.
5. AI use in UAVs, battlefield analytics.
6. Ethical use of AI in warefare and AI governance in defence applications

1. Introduction to AI in Neuroscience : An Overview

2. How Artificial Intelligence can help with important problems in Neuroscience.

3. Key Advantages of Integrating AI in Neuroscience

5. Applications of AI in Neuroscience applications

6. Real-World Use Cases of AI and Robotics Working Together

7. Hands-On Demonstration: AI in Neuroscience applications.

1. Feature Extraction from Different Modalities: Extract text embeddings (using BERT/GPT), image embeddings (using ResNet/Vision Transformer), and audio embeddings (using Wav2Vec). Compare different representation techniques and visualize them using t-SNE or PCA.
2. Implement a Transformer-based architecture for text generation and analyze the impact of hyperparameter tuning.
3. Extracting Audio Features using Librosa: Load an audio file and extract features like Mel spectrograms, MFCCs, and Chroma features. Visualize these features to understand how speech and music differ.
4. Fine-tune a pre-trained large language model (LLM) for text classification on a domain-specific dataset.
5. Concatenation-Based Feature Fusion: Take a text description of an image and its image features. Concatenate both and use a simple MLP classifier for classification tasks.
6. Implement an LLM for document summarization and evaluate its performance using ROUGE scores.
7. Compare Early vs. Late Fusion in Multimodal Classification: Take a dataset with text + images (e.g., Hateful Memes, VQA). Train two models:

a. Early fusion: Concatenate text + image features before classification.

b. Late fusion: Train separate classifiers and merge final predictions.

Evaluate both using Accuracy, F1-score, and AUC-ROC.

8. Use an LLM for machine translation between two languages and assess translation accuracy.
9. Image-to-Text Retrieval using CLIP: Load OpenAI’s CLIP model and input an image. Retrieve the most relevant text captions from a dataset. Try the reverse: Given a text description, retrieve the most similar image.
10. Develop a chatbot using an LLM and fine-tune it for task-oriented dialogue generation.
11. Evaluating Image Captioning Models: Generate captions using a pre-trained Image Captioning model.
12. Compare with ground-truth captions using:

BLEU, ROUGE, and METEOR (for text similarity).

BERTScore (for semantic similarity).

13. Implement a multimodal LLM for image and text-based question answering.
14. Explore the impact of various data augmentation techniques on the performance of multimodal AI models.
15. Explore prompt engineering techniques to optimize LLM responses for different applications.
16. Evaluating Speech-to-Text Models: Transcribe an audio file using Whisper or Wav2Vec. Compare output with ground truth using: Word Error Rate (WER) and Levenshtein Distance (edit distance).
17. Train a Reinforcement Learning with Human Feedback (RLHF) model to improve LLM-generated responses.
18. Text-Based Face Search: Use FaceNet or Dlib to extract facial embeddings. Input a text description like "smiling person with glasses" and rank images based on cosine similarity.
19. Use parameter-efficient fine-tuning methods (e.g., LoRA, Adapter Layers) to adapt an LLM to a specialized task.

20. Use a pre-trained VQA model to answer questions about images and evaluate its performance on a small dataset.
21. Implement an LLM-based code generation system and evaluate its ability to generate executable code.

Introduction to deep learning, CNNs, RNNs, LSTMs, Transformer architecture, Self-Attention mechanism, Multi-Head Attention, Vision Transformer (ViT), applications in image classification and object detection, Pretraining vs. Fine-Tuning paradigms, Transfer Learning techniques, and Fine-Tuning ViTs for custom tasks, understanding LLMs (GPT, BERT, T5), encoder-decoder architectures, Pretraining vs. Fine-Tuning in LLMs, Scaling Laws, Model Efficiency, Principles and techniques for fine-tuning pretrained models. 

Amity Centre for Artificial Intelligence (ACAI) and NVIDIA Deep Learning Institute (DLI) are organizing a hands-on workshop, exclusively for academic students and researchers.

Course Instructor will be Rakesh Chandra Joshi  who is a NVIDIA Certified Deep Learning Instructor and University Ambassador.

NVIDIA DLI certification will be provided to all the participants.

Theory: Deep Learning, Perceptron Model, Activation Function, Loss Function, Building Neural Networks, Cross Entropy Loss, Loss Optimization, Gradient Descent, Learning Rate, Overfitting, Dropouts,   Convolution Neural Networks: Feature Extraction, Feature Representation, Fully Connected Network, Convolution Layer & Feature Maps, Pooling, Activation Function, Transfer Learning and Fine-tuning, Popular CNN architectures such as AlexNet, VGG-Net, ResNet, GoogleNet, Introduction to RNNs and their applications, Long Short-Term Memory (LSTM) networks and Gated Recurrent Units (GRUs), Sequence-to-sequence models for natural language processing, Attention mechanism. Transformers.  

Hands-on Session:  Installation of software and IDE, Programming on Python, Intro to Training Resources, Import Packages and Loading of Datasets, Creating CNN Models from Scratch, Compiling of CNN Models, Training and Testing, Plotting of Curves,  Deep Learning Training and Architecture, Feature Extraction, Models training with some pre-trained models,   Text data handling, Train Recurrent Neural Networks (RNNs) on the dataset for sentiment analysis, Training an Attention Network for classification. Implementation of convolution attention block module.

Hands-on Training to configure edge devices (Like Raspberry Pi/Jetson Nano) and deploy AI models for real-world AI applications.

For Students who has experience to train Deep Learning models - Learn to deploy it in edge devices.

Theory: Machine Learning, Supervised Learning, Perceptron model, Building Neural Networks, Cross Entropy loss, loss optimization, Gradient decent, Learning rate, Overfitting, Dropouts, CNN, Case study of use of Business analytics, Machine learning for Business analytics, Case study of use of Deep Learning based AI models for Business analytics

Hands-on Session: Basic Python Programming, Intro to Training resources, Introduction to packages and loading of datasets, Creating CNN models from scratch for Business analytics, Compiling of models, Training and Testing of Models, Training with pertained models for business analytics.

Theory: Machine Learning, Supervised Learning, Perceptron model, Activation function, Loss Function, Building Neural Networks, Cross Entropy loss, loss optimization, Gradient decent, Learning rate, Overfitting, Dropouts, CNN, Case study of use of Machine learning for Biological / Clinical Applications, Case study of use of Deep Learning based CNN models for Biological / Clinical Applications. 

Hands-on Session: Basic Python Programming, Intro to Training resources, Introduction to packages and loading of datasets, Creating CNN models from scratch for Biological Application, Compiling of models, Training and Testing of Models, Training with pertained models for Biological / Medical Applications. 

Theory: Machine Learning, Supervised Learning, Perceptron model, Activation function, Loss Function, Building Neural Networks, Cross Entropy loss, loss optimization, Gradient decent, Learning rate, Overfitting, Dropouts. Features Extraction, Feature Representation, Fully Connected Neural Network, CNN, pooling, stride, padding, Transfer Learning and Fine-tuning.

Hands-on Session: Basic Python Programming, Intro to Training resources, Import packages and loading of datasets, Creating CNN models from scratch, Compiling of models, Training and Testing of Models, Training with pertained models.

Theory: Perceptron model, Activation function, Loss Function, Building Neural Networks, Cross Entropy loss, loss optimization, Gradient decent, Learning rate, Over fitting, Dropouts. Features Extraction, Feature Representation, Fully Connected Neural Network, CNN, pooling, stride, padding, Transfer Learning and Fine-tuning.Recurrent Neural Networks (RNNs), Sequential models for solving Long-term dependency issues. LSTM models, Introduction to Autoencoders and Variational Autoencoders, Generative Adversarial Networks (GANs), Applications of GANs.

Hands-on Session: Basic Python Programming, Intro to Training resources, Import packages and loading of datasets, Creating ANN & CNN models from scratch, Compiling of models, Training and Testing, Plotting of curves, feature extraction, Models training with some pertained models, Image Augmentation, Implementation of RNN, GAN architecture implementation.

2 weeks research project to build a Deep Neural Network for solving a problem. 

Theory: Perceptron model, Activation function, Loss Function, Building Neural Networks, Cross Entropy loss, loss optimization, Gradient decent, Learning rate, Over fitting, Dropouts. Features Extraction, Feature Representation, Fully Connected Neural Network, CNN, pooling, stride, padding, Transfer Learning and Fine-tuning.

Recurrent Neural Networks (RNNs), Sequential models for solving Long-term dependency issues. LSTM models, Introduction to Autoencoders and Variational Autoencoders, Generative Adversarial Networks (GANs), Applications of GANs.

Hands-on Session: Basic Python Programming, Intro to Training resources, Import packages and loading of datasets, Creating ANN & CNN models from scratch, Compiling of models, Training and Testing, Plotting of curves, feature extraction, Models training with some pertained models, Image Augmentation, Implementation of RNN, GAN architecture implementation. 

Theory: Perceptron model, Activation function, Loss Function, Building Neural Networks, Cross Entropy loss, loss optimization, Gradient decent, Learning rate, Overfitting, Dropouts. Features Extraction, Feature Representation, Fully Connected Neural Network, CNN, pooling, stride, padding, Transfer Learning and Fine-tuning.

Hands-on Session: Basic Python Programming, Intro to Training resources, Import packages and loading of datasets, Creating CNN models from scratch, Compiling of models, Training and Testing of Models, Training with pertained models.