Synthesizing Action Sequences for Modifying Model Decisions
arXiv:1910.00057 · doi:10.1609/aaai.v34i04.5996
Abstract
When a model makes a consequential decision, e.g., denying someone a loan, it needs to additionally generate actionable, realistic feedback on what the person can do to favorably change the decision. We cast this problem through the lens of program synthesis, in which our goal is to synthesize an optimal (realistically cheapest or simplest) sequence of actions that if a person executes successfully can change their classification. We present a novel and general approach that combines search-based program synthesis and test-time adversarial attacks to construct action sequences over a domain-specific set of actions. We demonstrate the effectiveness of our approach on a number of deep neural networks.
References in corpus (17)
- Adam: A Method for Stochastic Optimization
- Continuous control with deep reinforcement learning
- Explaining and Harnessing Adversarial Examples
- A Unified Approach to Interpreting Model Predictions
- Axiomatic Attribution for Deep Networks
- Towards Deep Learning Models Resistant to Adversarial Attacks
- Actionable Recourse in Linear Classification
- Neural Programmer-Interpreters
- COLIN: Planning with Continuous Linear Numeric Change
- Neuro-Symbolic Program Synthesis
- Programming with a Differentiable Forth Interpreter
- TerpreT: A Probabilistic Programming Language for Program Induction
- Leveraging Grammar and Reinforcement Learning for Neural Program Synthesis
- Interpreting Neural Network Judgments via Minimal, Stable, and Symbolic Corrections
- Differentiable Programs with Neural Libraries
- HOUDINI: Lifelong Learning as Program Synthesis
- Scalable Planning with Tensorflow for Hybrid Nonlinear Domains