Agent-based simulation + reinforcement learning is the topic of an industry-insider panel discussion at the AnyLogic Conference this April 17–18. Join us and learn how simulation modeling and continuously learning agents are shaping the future of reinforcement learning.
Artificial intelligence is everywhere. That’s what it feels like. If it isn’t currently employed for a task, then it is being talked of — in particular, combining machine learning with simulation.
Governments are in the mix as well. This light-hearted tweet [animation], evidence even lords have AI under their watchful eye.
Over the course of the past few months, we’ve continued to receive written evidence to our inquiry, including from @DeepMindAI, @LeverhulmeCFI, @NCCGroupplc & @GoogleUK. Read it all here: https://t.co/yxm9iQhdfc pic.twitter.com/RkRdBYXNCF— Lords AI Committee (@LordsAICom) January 11, 2018
Here at AnyLogic, we enable simulation modeling to solve real-world problems, safely and efficiently — an especially valuable ability when harnessing developments in AI. Lyle Wallis, a PwC Analytics Director, recently shared examples and experience from integrating AI and simulation modeling at their Artificial Intelligence Accelerator. Here we take a look at what was a standing room only presentation at the recent Winter Simulation Conference.
The Artificial Intelligence Accelerator at PwC is using AnyLogic simulation and other AI technologies in the creation of a new generation of simulation models. The aim of this work is to reflect the pervasive adoption of AI across business and society. So it follows, if AI is in the real world, simulation models must also adopt AI as well!
What is the effect of AI and simulation on business and industry?
The impact of AI is ‘game changing’, with applications ranging from automating existing processes to disrupting markets and business models. The Accelerator identifies three areas of impact:
Disrupting your core business
For example, process automation in accounting and law firms – A robot lawyer has already overturned 375,000 parking tickets.
Innovating with new services
The development of products and services such as Amazon’s Alexa or the Google Assistant. Companies apply analytics, machine learning and simulation modeling to their big data and enable new services for their customers.
Redesigning your business model
A reality being faced by car manufacturers. Early movers are offering vehicles and services to capture the car share/ride markets. BMW with DriveNow for instance.
Why use simulation with AI and how they will be integrated?
Systems that already use AI, such as digital supply chains, smart factories, and other industrial processes that form industry 4.0, will need to include AI in their simulation models. For example, with digital twins (AL intro paper) and what-if analysis systems, the AI components can be directly embedded in the simulation model to allow testing and forecasting.
Applying AI to optimization and calibration is another key opportunity in simulation modeling. Agent based systems often have a lot of parameters and to explore all their permutations can require impractical run times. Machine learning and intelligent sampling can be used to create meta-models that can deliver dramatic speed increases for large-scale agent based models.
In the case of deep learning, components can be developed to replace rule-based models. This is possible when considering human behavior and decision making. These deep learning components can either be used in simulation models to reflect the real system, or simulation models can be used to train the AI components. By generating the data sets necessary for neural network training, simulation models can be a powerful tool when deploying deep learning in the real world.
A case in point: Applying deep reinforcement learning to autonomous vehicle development
PwC is working with a large car company looking to introduce autonomous vehicles for the public. Part of this work employs deep reinforcement learning to develop rules. Together with simulation, deep reinforcement learning is used to determine ‘optimal’ decision rules that allow the vehicles to maximize efficiency while also satisfying customer trip demand.
The software environment for the project uses the extensible and practical environment of AnyLogic to lever the capabilities of DL4J for the deep learning environment.
Autonomous cars are becoming more common and the features are already in many consumer cars. AI becomes more pervasive in the real world with every project, and necessarily it must be part of our simulations. It will not only be part of our simulations, our simulations will also help develop the AI.
More details of this work, and further details describing the effects of AI on industry and business can be found in Lyle Wallis’s original PwC presentation.
Integrating Artificial Intelligence with Simulation Modeling presentation:
Are you using AI in your simulation models yet? Have you connected with the DL4J platform? Do you agree that pervasive AI in the real world means we must also have pervasive AI in our simulated world as well? Let us know in the comments below!
⚡ Lyle Wallis presented the latest developments and updates on this topic at the AnyLogic Conference, April 2018 - What links machine learning, deep learning, and simulation?