Even after lot off Low code and No Code platforms emerge in the form of Auto ML. (As machine learning has moved from theory to applications). Still Machine Learning has some hard lesion to follow which are
Here’s what we know to be true
- The more data, the better. Having a large, pre-existing repository of data governed by well-established rules is an absolute pre-requisite for teaching a computer. A minimum volume of data are needed for learning to occur, and after that, the more data, the better.
- Garbage in, garbage out. It’s trite, but true. Data must be clean and linked to be useful.
- Machines can’t do it alone. The only way for machines to learn is for a human teacher to review their work and to educate them on “right” from “wrong.” This iterative process improves the computer’s output. And the collaboration between man and machine must be sustained. Experts must be on hand at all times to validate what machines do, especially as data and data models are constantly changing.
- Domain knowledge rules. It is impossible to select the scientific algorithms, configure the systems, and evaluate the computer’s output without being intimately familiar with the subject matter. This requires years of experience in the specific domain. There is no crash course on this.
- There are no shortcuts. Training computers takes time and trial and error. Data investigations of possible errors on the machine’s part translate, over time, into cleaner data and better rules and algorithms. But the demands keep escalating over time. The level of training that was good enough today will be insufficient tomorrow.
- The only good algorithm is a specific algorithm. A machine trained on financial systems will produce poor results in healthcare. Even in healthcare, a machine trained on a specific data set will not perform on another data set. Algorithms must be specific to the use case, country, language, data type, and even regional healthcare delivery procedures.
Source:
https://www.cluzters.ai/vault/288/2960/how-machines-learn-in-healthcare
https://docs.google.com/file/d/1WCZ0KLZjz9vA6HcO2zeCZJDEq7jQqD9d/view
Hard lessons of machine learning
