From Functional Thinking to Machine Learning Engineering
Philippe Touati
Management Board | Corporate & Institutional Banking | Tech & Communications
The value of AI comes from predictive and prescriptive analytics not from from descriptive analytics or diagnostics. Belmond Capital Limited Patricia Poon Philippe Touati hashtag. The challenge is how you do that in the AI context.
Functional thinking
The traditional way, used to built expert system, is what we call functional thinking. where you bring expert in an industry, let's say an acoustic engineering or a chief risk officer and try to capture their knowledge and model it upfront. and then use data to validate your model.
Machine Learning Engineering
Machine learning requires a different approach. it tries to start with data after you choose a algorithmic model, let's say classification logistic regression or soft-max output with deep learning. This is the approach Netflix, Amazon, Go-Jek takes when they deploy new Algorithms
This approach is also engineering approach: Machine Learning Engineering.
You can see the flow below.
At Belmond Capital Limited, we believe that constant learning, research in AI business model with deep understanding of 3 industries (retail, financial services and telecoms) continuous coding and mathematics knowledge acquisition are critical to master current transition. Education is source of competitive advantage.
#analytics.. Are you AI ready?
Follow-us as www.BelmondCapital.com or at our linkedin page www.dhirubhai.net/company/belmond-capital-limited/
Chief Investment Officer at Noviscient | Financial Risk Management, Hedge Funds
5 年Right - interesting. The expert rules approach is too fragile - the system breaks if things change. As you point out, this has moved forward to having the machine learning build the models (rather than the human design the model) from the data. This works if you have big data. The challenge is what to do if you have small data - which is often the case in finance. What do you think? It feels like you need to provide some 'structure' to the machine learning models.
Director at Joyce A. Tan & Partners LLC
6 年Whether functional or machine learning, it seems to me that design is a common step in the process. One wonders how long this remains a human role