An official website of the United States government
Symbolic planning models allow decision-making agents to sequence actions in arbitrary ways to achieve a variety of goals in dynamic domains. However, they are typically handcrafted and tend to require precise formulations that are not robust to human error. Reinforcement learning (RL) approaches do not require such models, and instead learn domain dynamics by exploring the environment and collecting rewards. However, RL approaches tend to require millions of episodes of experience and often learn policies that are not easily transferable to other tasks. In this paper, we address one aspect of the open problem of integrating these approaches: how can decision-making agents resolve discrepancies in their symbolic planning models while attempting to accomplish goals? We propose an integrated framework named SPOTTER that uses RL to augment and support ("spot") a planning agent by discovering new operators needed by the agent to accomplish goals that are initially unreachable for the agent. SPOTTER outperforms pure-RL approaches while also discovering transferable symbolic knowledge and does not require supervision, successful plan traces or any a priori knowledge about the missing planning operator.
more »
« less
Network planning with deep reinforcement learning
Network planning is critical to the performance, reliability and cost of web services. This problem is typically formulated as an Integer Linear Programming (ILP) problem. Today's practice relies on hand-tuned heuristics from human experts to address the scalability challenge of ILP solvers. In this paper, we propose NeuroPlan, a deep reinforcement learning (RL) approach to solve the network planning problem. This problem involves multi-step decision making and cost minimization, which can be naturally cast as a deep RL problem. We develop two important domain-specific techniques. First, we use a graph neural network (GNN) and a novel domain-specific node-link transformation for state encoding, in order to handle the dynamic nature of the evolving network topology during planning decision making. Second, we leverage a two-stage hybrid approach that first uses deep RL to prune the search space and then uses an ILP solver to find the optimal solution. This approach resembles today's practice, but avoids human experts with an RL agent in the first stage. Evaluation on real topologies and setups from large production networks demonstrates that NeuroPlan scales to large topologies beyond the capability of ILP solvers, and reduces the cost by up to 17% compared to hand-tuned heuristics.
more »
« less
- Award ID(s):
- 1918757
- PAR ID:
- 10341119
- Date Published:
- Journal Name:
- Proceedings of the 2021 ACM SIGCOMM 2021 Conference
- Page Range / eLocation ID:
- 258 to 271
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Abstract The crowdsourcing literature has shown that domain experts are not always the best solvers for complex system design problems. Under certain conditions, novices and specialists in adjacent domains can provide novel solutions at lower costs. Additionally, the best types of solvers for different problems are dependent on the architecture of complex systems. The joint consideration of solver assignment and system decomposition, referred to as solver-aware system architecting (SASA), expands traditional system architecting practices by considering solver characteristics and contractual incentive mechanisms in the design process and aims to improve complex system design and innovation by leveraging the strengths of domain experts, crowds, and specialists for different parts of the problem. The joint consideration of problem decomposition and solver assignment decisions in SASA renders the design space exponentially more complex. Therefore, new computationally efficient and mathematically rigorous methods are needed to explore this high-dimensional space and extract reliable heuristics. To address this need, this paper presents a computational approach using a Markov decision process (MDP) formulation, Q-learning, and Gaussian mixture models. Together, these techniques explore the large space of possible solver–module assignments by modeling the sequential nature of solver assignment decisions, capturing these temporal dependencies, thereby enabling optimization for long-term expected rewards, and analyzing reward distributions. The approach identifies heuristics for solver assignment based on the designer’s preference for cost-performance trade-off through the parameterized reward function. The approach is demonstrated using a simple and idealized golf problem, which has characteristics similar to design problems, including how the problem is decomposed into interdependent modules and can be solved by different solvers with different strengths that interact with the module type. The results show that the proposed approach effectively elicits a rich set of heuristics applicable in various contexts for the golf problem and can be extended to more complex systems design problems.more » « less
-
Tasks across diverse application domains can be posed as large-scale optimization problems, these include graphics, vision, machine learning, imaging, health, scheduling, planning, and energy system forecasting. Independently of the application domain, proximal algorithms have emerged as a formal optimization method that successfully solves a wide array of existing problems, often exploiting problem-specific structures in the optimization. Although model-based formal optimization provides a principled approach to problem modeling with convergence guarantees, at first glance, this seems to be at odds with black-box deep learning methods. A recent line of work shows that, when combined with learning-based ingredients, model-based optimization methods are effective, interpretable, and allow for generalization to a wide spectrum of applications with little or no extra training data. However, experimenting with such hybrid approaches for different tasks by hand requires domain expertise in both proximal optimization and deep learning, which is often error-prone and time-consuming. Moreover, naively unrolling these iterative methods produces lengthy compute graphs, which when differentiated via autograd techniques results in exploding memory consumption, making batch-based training challenging. In this work, we introduce ∇-Prox, a domain-specific modeling language and compiler for large-scale optimization problems using differentiable proximal algorithms. ∇-Prox allows users to specify optimization objective functions of unknowns concisely at a high level, and intelligently compiles the problem into compute and memory-efficient differentiable solvers. One of the core features of ∇-Prox is its full differentiability, which supports hybrid model- and learning-based solvers integrating proximal optimization with neural network pipelines. Example applications of this methodology include learning-based priors and/or sample-dependent inner-loop optimization schedulers, learned with deep equilibrium learning or deep reinforcement learning. With a few lines of code, we show ∇-Prox can generate performant solvers for a range of image optimization problems, including end-to-end computational optics, image deraining, and compressive magnetic resonance imaging. We also demonstrate ∇-Prox can be used in a completely orthogonal application domain of energy system planning, an essential task in the energy crisis and the clean energy transition, where it outperforms state-of-the-art CVXPY and commercial Gurobi solvers.more » « less
-
Recent successes of Reinforcement Learning (RL) allow an agent to learn policies that surpass human experts but suffers from being time-hungry and data-hungry. By contrast, human learning is significantly faster because prior and general knowledge and multiple information resources are utilized. In this paper, we propose a Planner-Actor-Critic architecture for huMAN-centered planning and learning (PACMAN), where an agent uses its prior, high-level, deterministic symbolic knowledge to plan for goal-directed actions, and also integrates the Actor-Critic algorithm of RL to fine-tune its behavior towards both environmental rewards and human feedback. This work is the first unified framework where knowledge-based planning, RL, and human teaching jointly contribute to the policy learning of an agent. Our experiments demonstrate that PACMAN leads to a significant jump-start at the early stage of learning, converges rapidly and with small variance, and is robust to inconsistent, infrequent, and misleading feedback.more » « less
-
Collective, especially group-based, managerial decision making is crucial in organizations. Using an evolutionary theoretic approach to collective decision making, agent-based simulations were conducted to investigate how human collective decision making would be affected by the agents’ diversity in problem understanding and/or behavior in discussion, as well as by their social network structure. Simulation results indicated that groups with consistent problem understanding tended to produce higher utility values of ideas and displayed better decision convergence, but only if there was no group-level bias in collective problem understanding. Simulation results also indicated the importance of balance between selection-oriented (i.e., exploitative) and variation-oriented (i.e., explorative) behaviors in discussion to achieve quality final decisions. Expanding the group size and introducing nontrivial social network structure generally improved the quality of ideas at the cost of decision convergence. Simulations with different social network topologies revealed collective decision making on small-world networks with high local clustering tended to achieve highest decision quality more often than on random or scale-free networks. Implications of this evolutionary theory and simulation approach for future managerial research on collective, group, and multilevel decision making are discussed.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript
- Conference Paper:
- https://doi.org/10.1145/3452296.3472902
