AutoAI
Putting Systems at the Heart of Machine Learning
Neil D. Lawrence
2019-10-30
10 Minute Talk, Wednesday Meeting, Department of Computer Science and Technology, University of Cambridge
Project Description
It used to be true that computers only did what we programmed them to do, but today AI systems are learning from our data. This introduces new problems in how these systems respond to their environment.
We need to better monitor how data is influencing decision making and take corrective action as required.
Aim
Scale safe and reliable AI solutions.
Move from Auto ML to Auto AI
Bayesian Optimisation to Bayesian System Optimisation
SafeBoda
With road accidents set to match HIV/AIDS as the highest cause of death in low/middle income countries by 2030, SafeBoda’s aim is to modernise informal transportation and ensure safe access to mobility.
Turing AI Fellowship
and
Your Name Here
Inclusive Project
There is no way that the team we’re building will be able to deliver on this agenda alone, so please join us in addressing these challenges!
Ride Allocation Prediction
The Promise of AI
Automation forces humans to adapt, we serve.
We can only automate by systemizing and controlling environment.
AI promises to be first wave of automation that adapts to us rather than us to it.
That Promise …
… will remain unfulfilled with current systems design.
Computer Science Paradigm Shift
Von Neuman Architecture:
Code and data integrated in memory
Today:
Code and data separated for security
Computer Science Paradigm Shift
Machine learning:
Machine learning is a high level breach of the code/data separation.
Technical Consequence
Classical systems design assumes decomposability .
Data-driven systems interfere with decomponsability.
Bits and Atoms
The gap between the game and reality.
The need for extrapolation over interpolation.
Data Oriented Architectures
Convert data to a first-class citizen .
View system as operations on data streams .
Expose data operations in a programmatic way.
Streaming System
Move from pull updates to push updates.
Operate on rows rather than columns.
Lead to stateless logic: persistence handled by system.
Example Apache Kafka + Apache Flink
Hypothetical Streams
Real stream — share prices
derived hypothetical stream — share prices in future.
Hypothetical constrained by
input constraints.
decision functional
computational requirements (latency)
Hypothetical Advantage
Modelling is now required.
But modelling is declared in the ecosystem.
If it’s manual, warnings can be used
calibration, fairness, dataset shift
Opens door to auto-adaptable ML.
Ride Sharing: Service Oriented
Ride Sharing: Data Oriented
Ride Sharing: Hypothetical
Auto AI
Auto ML is great but not sufficient
Interacting components in an ML system
Identify problems, and automatically deploy solutions
Conclusion
Paradigm shift for computer science.
Want to study deployed interacting ML systems.
Need to put the data flows at the heart, not models or services.
Need expertise in
Security, Programming languages, Systems
Implications for Hardware and network design