NeurIPS 2019 Expo Workshop

Dec. 8, 2019

Expo 2019 Schedule »

Causal Inference & Reinforcement learning: Making the right intervention

Sponsor: QuantumBlack

Abstract:

The successes of predictive modelling encourage many organizations to leverage machine learning models in their decision making. Examples include the design of marketing, and user relationship strategies. Such analyses typically pose the following challenges: (1) the need to distinguish between events that cause outcomes from those that merely correlate; (2) the need to optimize the medium to long term results, instead of focussing on short term gains alone. The two areas of ML research that are particularly relevant here are Causal Inference (CI) and Reinforcement learning (RL).

There have been astonishing successes in the fields of RL and CI over recent years. However, applications in industry are lagging behind. One key challenge is to choose the right analytical framing to make the approach tractable, whilst also solving the real-world problem. Moreover, existing applications often assume the ability to perform experiments, which is infeasible for many questions of interest in industry. Models have to rely on historically observed data alone. Finally, the lack of mature open source software makes attempts to apply RL and CI in fast-paced environments risky.

In this workshop, we will provide an overview of QB’s experiences implementing these methods. Our objective is to facilitate a discussion about practical challenges in real-world applications: 1. Introduction 2. Framework for shaping real-world strategy problem into ML problem 3. CI in practice with focus on Bayesian Networks 4. RL in practice with focus on offline learning 5. Closing remarks and discussion

At QB, we're passionate about diversity. We are actively working to achieve more equality in tech in all dimensions including gender, race, nationality, and sexual orientation. It is important to us that the presenters of this workshop will reflect our diversity. Some recent articles we wrote on gender diversity are tiny.cc/1zlucz and tiny.cc/p1lucz, and we are proud sponsors of WiML.