Perspectives

AI in Equipment and Auto Finance

Part 3: Moving Forward with Machine Learning

Produced in association with Alfa iQ, Moving Forward with Machine Learning is Alfa's third and final paper on artificial intelligence in the asset finance industry. It combines the theoretical insights from our position paper Balancing Risk and Reward with the use cases explored in our technical paper Using Machine Learning in the Wild.

Moving Forward explores the trajectory of machine learning, its uses in auto and equipment finance, and how ML will continue to advance in the near future. There follows an in-depth exploration of federated learning, and how organisations can use private data to train ML models without ever compromising the privacy of that data.

The meteoric rise of Machine Learning

Why have we seen so many breakthroughs that make ML relevant for application, such as the production of autonomous vehicles, and ML models that can play games like Go and chess better than humans can?
The meteoric rise of Machine Learning

Improving ML models requires not just more data, but data that’s clean, well structured, and representative of a wider variety of real-world situations.
The meteoric rise of Machine Learning

Along with the dramatic increases in the volume of data we're capturing, the quality of that data is also increasing.

Federated learning aims to bring the "learning" part of the process to the data owner, avoiding the need to transfer any private data.

Achieving federated learning through differential privacy. Using Alfa Systems data as an example, the diagram below shows how multiple data owners can contribute to the training of a single ML model, without sharing data.

Achieving federated learning through differential privacy

An initial ML model is deployed in the same environment as Alfa Systems, alongside an ML application.
Alfa Systems APIs are invoked to extract the data on which to train the ML model.
Random “noise” is added to the updates before sending them outside the environment's network.
The update is sent to an impartial third party.
The updated ML model is redistributed to the data owners.

Data privacy and federated learning

Data’s rapid rate of growth will continue. Meanwhile, one of its important associated qualities is decreasing: accessibility.
Data privacy and federated learning

The most useful data for training an ML model is, more often than not, private.
Data privacy and federated learning

Federated learning allows data owners to contribute to ML training without sharing their private data.
Data privacy and federated learning

Some techniques like differential privacy - which can be used in combination with federated learning - come with a trade-off between privacy and the resulting accuracy of the ML model.

A partnership between Alfa and Bitfount, Alfa iQ was established to deliver intelligence to the world’s auto and equipment finance providers, with a mission to make access to assets efficient, intelligent and fair.

Alfa iQ will provide industry-leading direction in the use of ML. The partnership brings together Alfa’s industry experience and asset finance data models, with Bitfount’s expert AI data scientists and federated learning platform. Its ambition is to provide the best machine learning models and advanced decisioning scorecards for the asset finance industry.

AlfaIQ_Logo_Sm_210630_154932.png?mtime=20210630154932#asset:3790

Alfa iQ: The intelligence platform for asset finance. Learn more and stay updated at alfaiq.com.

The future of Machine Learning

ML is a branch of technology that will only increase in popularity and mainstream adoption - even in more risk-averse industries like auto and equipment finance.
The future of Machine Learning

Models which until recently would have required a supercomputer to train can now achieve similar results when trained on laptops.
The future of Machine Learning

Numerous challenges remain, particularly in areas such as ethics, bias, fairness, data quality and interpretability. However, none of these is insurmountable.

Document

AI in Equipment and Auto Finance Part 3: Moving Forward with Machine Learning
01/07/2021

Perspectives

AI in Equipment and Auto Finance

Part 3: Moving Forward with Machine Learning

The meteoric rise of Machine Learning

Why have we seen so many breakthroughs that make ML relevant for application, such as the production of autonomous vehicles, and ML models that can play games like Go and chess better than humans can?

The meteoric rise of Machine Learning

Improving ML models requires not just more data, but data that’s clean, well structured, and representative of a wider variety of real-world situations.

The meteoric rise of Machine Learning

Along with the dramatic increases in the volume of data we're capturing, the quality of that data is also increasing.

Achieving federated learning through differential privacy

An initial ML model is deployed in the same environment as Alfa Systems, alongside an ML application.

Alfa Systems APIs are invoked to extract the data on which to train the ML model.

Random “noise” is added to the updates before sending them outside the environment's network.

The update is sent to an impartial third party.

The updated ML model is redistributed to the data owners.

Data privacy and federated learning

Data’s rapid rate of growth will continue. Meanwhile, one of its important associated qualities is decreasing: accessibility.

Data privacy and federated learning

The most useful data for training an ML model is, more often than not, private.

Data privacy and federated learning

Federated learning allows data owners to contribute to ML training without sharing their private data.

Data privacy and federated learning

Some techniques like differential privacy - which can be used in combination with federated learning - come with a trade-off between privacy and the resulting accuracy of the ML model.

The future of Machine Learning

ML is a branch of technology that will only increase in popularity and mainstream adoption - even in more risk-averse industries like auto and equipment finance.

The future of Machine Learning

Models which until recently would have required a supercomputer to train can now achieve similar results when trained on laptops.

The future of Machine Learning

Numerous challenges remain, particularly in areas such as ethics, bias, fairness, data quality and interpretability. However, none of these is insurmountable.

Document

AI in Equipment and Auto Finance Part 3: Moving Forward with Machine Learning

AI in Equipment and Auto Finance Part 3: Moving Forward with Machine Learning

AI in Equipment and Auto Finance Part 2: Using Machine Learning in the Wild

LIBOR Reform: The Impact for Auto and Equipment Finance Providers