I would like to start by highly emphasizing the following course from Udemy: The Product Management for AI & Data Science Course
It is giving a great beginners all round picture about AI and AI projects. Here are some 5 minutes read takeaways to inspire you to check the course:
AI projects should be viewed as R&D projects
Unlike typical software projects where we are used to more exact Scope, Timelines and Cost expectation, with AI projects that is harder the case. What represent them much more are:
- uncertain timeline
- uncertain budget
- uncertain result precision
The areas where AI is used are:
- Natural language processing
- Computer vision
- Speech recognition and
- Robotics
Types of AI
Machine learning uses statistics.
Deep Learning uses artificial neural networks.
Deep learning is more expensive, so with little amount of data use ML as it is cheaper and gives equal results
Supervised learning – show the machine what you want it to do
Unsupervised learning – models learn on their own without the need oof labeled data by finding patterns and commonalities
Supervised learning works best when you know what the outcome would be.
Unsupervised learning works best when you don’t know what the outcome will be
Reinforcement learning is when you give the end goal and set of rules, but no labeled data
When should AI be used
- When a human expert can give response in few seconds
- When it is hard to define exact rules of the problem solving
- When it is easy to get examples of the desired behavior – get labeled data
The AI Canvas (questions to answer)
- PREDICTION – What do you need to know to make the decision?
- JUDGEMENT – How do you value different outcomes and errors?
- ACTION – What are you trying to do?
- OUTCOME – What are your metrics for task success?
- INPUT – What data do you need to run the predictive algorithm?
- TRAINING – What data do you need to train the predictive algorithm?
- FEEDBACK – How can you use the outcomes to improve the algorithm?
Data collection methods
Next to a proper model second most important thing for AI project to succeed is the data the we train the model with.
These are the data collection methods:
- Open data
- Company data
- Crowdsourcing labeled data
- New feature data
- Acquisition or purchased data
There was also introduced a concept of MVD which stands for Minimum Viable Data (needed to train the model in order to develop an AI product)
Managing AI projects
An example of Kanban board states for an AI project:
- Data Backlog – when Datasets are ready
- Data Processing – preparing and cleaning the Data
- Data Modeling – building Models with clean data
- Model Training – train and retrain models to reach requirements
- Testing – test the trained Model
- Deployment – deploying the Model on production environment
- Monitoring – monitoring Model performance
- Done – celebration 🙂
Data & AI team roles
- Data/AI Product Manager – determines which products to build
- Data Scientist – builds AI models and conducts advanced data analysis
- ML Data Engineer – develops and maintains the ingestion of data, data infrastructure, data warehousing and data pipeline
The End
I hope you enjoyed the summary of this course and that it has inspired to you to check it out. It was very well worth it for understanding the basics of AI projects.
Which course or learning material would you like to share with me?
Connect with me on LinkedIn and let me know in a message https://vladimirrodic.com/about-me/