Invited as a speaker for Goa’s 1st ever Developer Conference 2022 organized by Goa Chamber of Commerce and Industry.
Note: The talk was before the announcement of openai chatgpt based models.
Here is the summary what was covered, the speaker discusses various aspects of AI/ML, including supervised and unsupervised learning, neural networks, data modalities, model training, and model size reduction techniques. They emphasize the importance of data in decision-making for AI/ML, and highlight the potential for AI to automate tasks that typically require significant human effort and creativity. The speaker concludes by acknowledging the momentum in the field of AI/ML, and the numerous opportunities for its application in various sectors.
Link to presentation
- https://tinyurl.com/ai-ml-presentation
- 00:00:00 In this section of the video, the speaker explains how the approach of AI/ML differs from standard programming. Using the example of identifying number plates of cars and determining their respective state, he describes how standard programming involves writing multiple functions and conditional statements, while AI/ML involves feeding the model the images and the associated label, allowing it to learn and map the input to the output. He also provides an overview of the different types of machine learning tasks: supervised, unsupervised, and reinforcement learning. The latter is demonstrated with the example of a game where the AI learns by trial and error with the only specification being a reward system.
- 00:05:00 In this section, the speaker discusses two techniques used in AI/ML: supervised and self-supervised learning. Self-supervised learning utilizes the inherent structure of the data to train models, assigning probability values to certain words based on context. The speaker goes on to explain the terms AI, machine learning, and deep learning, with deep learning being a subset of machine learning that utilizes neural networks for mapping input to output. The speaker then introduces the concept of a convolution neural network and a recurrent neural network and their applications in processing different types of data. Finally, they discuss the scope of AI/ML systems, emphasizing that deep learning requires significant amounts of data and opportunities for application exist in every sector with available data, such as cyber security, healthcare, and finance.
- 00:10:00 In this section, the speaker discusses the importance of data and its various modalities for building AI/ML models. They explain that data plays a critical role in making decisions while modeling AI/ML problems and that the neural network architecture and hyperparameters are based on the kind of data available. The speaker also emphasizes that having more data is ideal for better model training and fine-tuning, and transfer learning can help in cases where specific data is required. They also mention the importance of building retraining pipelines for deploying models and the various techniques for reducing model size, such as pruning, model distillation, and quantization, which are often not talked about in blog posts or YouTube videos.
- 00:15:00 In this section, the speaker explains an example of how AI can generate NFTs using only a simple prompt. By providing a basic prompt of a frustrated slot attempting to fix a computer bug, the AI is able to generate unique NFTs with different facial expressions from the same prompt. The speaker emphasizes the potential of AI in automating tasks that typically require a lot of human effort and creativity, such as creating high-quality images or finding a cure for cancer. The video ends with the speaker thanking the audience and noting that there will be future momentum in the field of AI and ML.
Credits:
Thank you, Melroy, for capturing the video content in timestamped text.
