A Deep Reinforcement Learning Framework with Formal Verification

What is it about?

This paper presents a deep reinforcement learning (DRL) framework called "Multi State-Actor" (MuStAc) for constructing a valid and error-free career agent for predicting relevant and valid career steps for employees based on their profiles and company pathways. The framework is built on a combination of a human resources ontology and an Event-B model, which generates action spaces with respect to formal properties. The Event-B model and formal properties are derived using ontology transformation to B-machines. The training for the MuStAc framework is conducted in two phases: first, each actor is trained separately to perform well for its corresponding state, and then all actors are trained simultaneously, with each actor aiming to maximize the common expected return. The authors demonstrate the effectiveness of the MuStAc framework through a case study involving career planning in a simulated job galaxy. They also discuss the limitations and potential future work for the framework.

Featured Image

Photo by Markus Spiske on Unsplash

Photo by Markus Spiske on Unsplash

Why is it important?

This work is important because it addresses the need for ensuring the quality, integrity, and correctness of artificial intelligence (AI) and analytical models that are used in human resources (HR) management and talent development. As AI and data become increasingly prevalent in these areas, it is important to have systems in place to oversee and ensure that they are operating ethically and fairly, and that they do not have unintended consequences on the workforce. The MuStAc framework presented in this paper aims to provide a way to construct valid and error-free career agents that can make predictions about relevant and valid career steps for employees based on their profiles and company pathways. By using a combination of a HR ontology and formal verification methods, the authors aim to provide a way to ensure that the career agent's recommendations are based on accurate and reliable data and that they are in line with ethical and fair principles.

Perspectives