Toward Interpretable-AI Policies Using Evolutionary Nonlinear Decision Trees for Discrete-Action Systems

Dhebar, Yashesh; Deb, Kalyanmoy; Nageshrao, Subramanya; Zhu, Ling; Filev, Dimitar

doi:10.1109/tcyb.2022.3180664

Cited by 10 publications

(3 citation statements)

References 23 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The second relates to the passive use of algorithms in running utilities and services, introducing the concept of a policy designed by induction. For instance, methods such as Deep Reinforcement Learning are increasingly being used to find optimal policies for a given control task (e.g., Dhebar et al 2020;Behzadan and Munir 2017).…”

Section: Figure 1 Policy Cycle Analytics Tools (Acronym: Operational ...mentioning

confidence: 99%

Supporting Decision-Making within the Policy Cycle

Pluchinotta¹,

Daniell²,

Tsoukiàs³

2022

The Routledge Handbook of Policy Tools

View full text Add to dashboard Cite

The policy cycle is a model of interactive stages for structuring policy problems and making decisions to deal with them. The model encourages cycling through discrete activities, using different policy tools. In order to support decision-making within a policy cycle, a decisionaiding perspective is needed; namely formal tools and methods introduced to help decisionmakers develop reasoned decisions during the different phases of the policy-making process.The intention of this chapter is firstly, to introduce existing Decision Analysis analytic methods and tools that can be used to support the different phase of the policy cycle, using concepts from Decision Analysis and Policy Analytics. Secondly, the decision-aiding perspective for the design and evaluation of policies is introduced and discussed. Lastly, the chapter focuses on the importance of the policy design phase, introducing an innovative participatory tool to support policy makers and relevant stakeholders during the design of policy alternatives

show abstract

Section: Figure 1 Policy Cycle Analytics Tools (Acronym: Operational ...mentioning

confidence: 99%

Supporting Decision-Making within the Policy Cycle

Pluchinotta¹,

Daniell²,

Tsoukiàs³

2022

The Routledge Handbook of Policy Tools

View full text Add to dashboard Cite

show abstract

“…Dhebar et al [6] propose an evolutionary approach for producing DTs with nonlinear splits (i.e., hyperplanes defined by conditions) for reinforcement learning tasks. This approach is able to obtain very good performance in the tested tasks.…”

Section: Interpretable Aimentioning

confidence: 99%

Interpretable AI for policy-making in pandemics

Custode

Iacca

2022

Proceedings of the Genetic and Evolutionary Computation Conference Companion

View full text Add to dashboard Cite

Since the first wave of the COVID-19 pandemic, governments have applied restrictions in order to slow down its spreading. However, creating such policies is hard, especially because the government needs to trade-off the spreading of the pandemic with the economic losses. For this reason, several works have applied machine learning techniques, often with the help of special-purpose simulators, to generate policies that were more effective than the ones obtained by governments. While the performance of such approaches are promising, they suffer from a fundamental issue: since such approaches are based on black-box machine learning, their real-world applicability is limited, because these policies cannot be analyzed, nor tested, and thus they are not trustable. In this work, we employ a recently developed hybrid approach, which combines reinforcement learning with evolutionary computation, for the generation of interpretable policies for containing the pandemic. These policies, trained on an existing simulator, aim to reduce the spreading of the pandemic while minimizing the economic losses. Our results show that our approach is able to find solutions that are extremely simple, yet very powerful. In fact, our approach has significantly better performance (in simulated scenarios) than both previous work and government policies.

show abstract

“…The application of state division serves as an intuitive and powerful method for interpreting DRL policies, which currently functions as an auxiliary tool for mimic learning and explicating policy performance. In the context of mimic learning, (Soares et al 2021;Dhebar et al 2022;Liu et al 2023) visualize state-action patterns to refine approximations. For instance, (Liu et al 2023) define critical experience points around decision boundaries, establishing their significance in interpretable policy distillation.…”

Section: Introductionmentioning

confidence: 99%

Analyzing Generalization in Policy Networks: A Case Study with the Double-Integrator System

Zhang,

Han,

et al. 2024

AAAI

View full text Add to dashboard Cite

Extensive utilization of deep reinforcement learning (DRL) policy networks in diverse continuous control tasks has raised questions regarding performance degradation in expansive state spaces where the input state norm is larger than that in the training environment. This paper aims to uncover the underlying factors contributing to such performance deterioration when dealing with expanded state spaces, using a novel analysis technique known as state division. In contrast to prior approaches that employ state division merely as a post-hoc explanatory tool, our methodology delves into the intrinsic characteristics of DRL policy networks. Specifically, we demonstrate that the expansion of state space induces the activation function $\tanh$ to exhibit saturability, resulting in the transformation of the state division boundary from nonlinear to linear. Our analysis centers on the paradigm of the double-integrator system, revealing that this gradual shift towards linearity imparts a control behavior reminiscent of bang-bang control. However, the inherent linearity of the division boundary prevents the attainment of an ideal bang-bang control, thereby introducing unavoidable overshooting. Our experimental investigations, employing diverse RL algorithms, establish that this performance phenomenon stems from inherent attributes of the DRL policy network, remaining consistent across various optimization algorithms.

show abstract

Toward Interpretable-AI Policies Using Evolutionary Nonlinear Decision Trees for Discrete-Action Systems

Cited by 10 publications

References 23 publications

Supporting Decision-Making within the Policy Cycle

Supporting Decision-Making within the Policy Cycle

Interpretable AI for policy-making in pandemics

Analyzing Generalization in Policy Networks: A Case Study with the Double-Integrator System

Contact Info

Product

Resources

About