Schedule

Hosted by: Pearl Mary S, Ramachandra Kaladharasarguda, Kameshwar Rao Jammalamadaka Venkata, Sneha Deepak Jadhav

Overview:

Artificial General Intelligence (AGI) stands as a theoretical pinnacle in the realm of AI, representing systems capable of emulating human-like cognitive abilities, including reasoning, learning, and consciousness. The advent of technologies like AutoGPT has prompted discussions on whether it signifies a nascent form of AGI. We utilize AutoGPT to create a self-optimizing MLOps framework, enabling autonomous reasoning and decision-making within workflows for improved efficiency and performance.

Automating ML Workflow with AutoGPT:

The MLOps pipeline ensures seamless transition of models from development to production environments. Our in-house MLOps framework named “Dyna.MLOPs” encompasses key stages such as data collection, preprocessing, model training, evaluation, and deployment, all while emphasizing CI/CD principles. It promotes collaboration, supports multi-cloud environments, and features a modular architecture. Within the Dyna.MLOps platform, we initiate the task of onboarding ML workflows using AutoGPT. Leveraging the power of state-of-the-art Large Language Models (LLMs) such as GPT-3.5 and GPT-4, AutoGPT functions as a semi-autonomous agent interfacing with web applications and services to execute tasks autonomously, guided by user-defined goals. We invoke AutoGPT to undertake a series of steps, from data preprocessing to model evaluation, streamlining the process and enhancing efficiency.

Additionally, we use MLflow for continuous integration (CI), ensuring systematic logging of hyper-parameters, prompts, outputs and the intermediate python files for traceability and reproducibility.

Enhancing Heart Disease Prediction with MLOps Workflow and AGI:

We initiate the automation process by providing an initial prompt that outlines the AI role and objectives. AutoGPT effortlessly takes over from there, leveraging this defined framework to initiate and orchestrate the automation process. The process is outlined below.

•  Iterative Interaction with LLMs: AutoGPT engages in iterative interactions with LLMs i.e. GPT3.5 turbo/GPT 4 to generate responses and execute commands. This iterative process enables AutoGPT to refine its understanding and adapt its actions accordingly.
• Seamless Integration with Resources: AutoGPT seamlessly integrates with online resources and local applications relevant to the ML workflow. It accesses data sources, software tools, and services necessary for executing the specified tasks.
• Agile Traversal of ML Workflow Landscape: Leveraging its agility and precision, AutoGPT navigates through the various stages of the ML workflow. It progresses from data preprocessing to model evaluation, orchestrating each step with efficiency.
• Comprehensive Tracking with MLflow: Hyperparameters, prompts, outputs, and intermediate files.
• Automated Model Selection with High Accuracy: AutoGPT identified a Random Forest classifier achieving 99% accuracy for heart disease prediction, which is better than the model selection performed by human experts, replacing traditional trial-and-error methods.
• Enhanced MLOps Automation: Future advancements could enable the automation of MLOps CD pipelines using custom agents.