Machine learning predicts with cautious mind to avoid big losses
Credal Machine Learning for Risk-Averse Decision Making
Machine Learning
Summary
Machine learning models usually try to be good on average, but sometimes they can make very bad predictions that cause big problems. The authors propose a way to make these models more careful by considering not only the average case but also the worst cases using special mathematical tools called credal sets, which represent uncertainty more broadly. This method helps the model avoid dangerous mistakes even when it's not sure about the exact risks. They tested their approach on different tasks and found it keeps the model safe without losing much overall performance.
What this means in practice
- •For financial risk managers: Use credal set-based predictors to minimize possible financial losses when market conditions are uncertain or changing.
- •For autonomous vehicle engineers: Implement risk-aware decision rules to prevent catastrophic errors during unexpected driving scenarios or sensor failures.
Authors
Timo Löhr, Paul Hofman, Maximilian Muschalik, Eyke Hüllermeier
Abstract
In many machine learning applications, it is necessary to guard against worst-case scenarios and predictions that could result in substantial losses. In principle, this can be achieved by training risk-averse predictive models that minimize loss functions such as conditional value-at-risk (CVaR), rather than relying on models that perform well on average. In practice, however, the effectiveness of this approach to risk aversion is undermined by the learner's uncertainty regarding the true loss distribution and, consequently, the true CVaR. To achieve reliable risk-aversion, we propose a method in which this (epistemic) uncertainty is represented in terms of credal sets, i.e., sets of probability distributions. More specifically, we develop an efficient yet reliable learner that produces predictions in the form of credal sets and combine it with a novel decision rule that maps each credal set to a single predictive distribution for CVaR minimization. Across classification, under distribution shift, and in reinforcement learning, our approach reliably avoids catastrophic decisions, while sacrificing little in expected performance.