The Mercury Machine Learning Lab centers its research on three core themes: development and applications of artificial intelligence to the specific domain of online travel booking and recommendation service systems.
To drive these advancements, the lab leverages key technical AI components, focusing primarily on:
The Mercury Machine Learning Lab is dedicated to the development and applications of artificial intelligence to the specific domain of online travel booking and recommendation service systems.
The lab’s mission and vision are to cover fundamental research topics, ranging from model-based exploration, parallel model-based reinforcement learning, methods for combined online and offline evaluation, prediction methods that correct for undesired feedback loops and selection bias, domain generalization and domain adaptation, and novel language processing models for better generalization.
The impact of the lab lies in scientific advancements, as well as immediate implications in terms of practical relevance for modern online businesses like Booking that aim to maximize customer satisfaction in quickly changing markets with the help of sophisticated data analytics.
Model-based Exploration: Effectively performing exploration in the non-stationary and multi-faceted environments that Booking.com interacts with.
Bridging online and offline evaluation: To develop and evaluate methods, both theoretically and experimentally, that bridge the gap between online evaluation and offline (off-policy) evaluation.
Novel language processing models for better generalisation: To develop methods for training NLP models that explicitly target generalization across multiple related tasks.
CLAX: Fast and Flexible Neural Click Models in JAX Journal Article
In: 2026, (SIGIR 2026).
Testing Partially Identifiable Causal Queries using Ternary Tests Miscellaneous
2026, (UAI 2026).
Truthful or Fabricated? Using Causal Attribution to Mitigate Reward Hacking in Explanations Journal Article
In: 2025, (ICLR (2026)).
Unidentified and Confounded? Understanding Two-Tower Models for Unbiased Learning to Rank (Extended abstract) Journal Article
In: 2025, (CONSEQUENCES Workshop at RecSys (2025)).
Truthful or Fabricated? Using Causal Attribution to Mitigate Reward Hacking in Explanations Journal Article
In: 2025, (Workshop on Actionable Interpretability@ICML2025).
Conditional Forecasts and Proper Scoring Rules for Reliable and Accurate Performative Predictions Journal Article
In: 2025, (NeurIPS (2025)).
Unidentified and Confounded? Understanding Two-Tower Models for Unbiased Learning to Rank Journal Article
In: 2025, (ICTIR 2025).
Truthful or Fabricated? Using Causal Attribution to Mitigate Reward Hacking in Explanations Journal Article
In: 2025, (COLM (2025)).
Revisiting the Berkeley Admissions data: Statistical Tests for Causal Hypotheses Journal Article
In: 2025, (UAI (2025)).
Are Bayesian networks typically faithful? Journal Article
In: 2025, (Bernoulli).
On Selection Bias in Statistical Causality Journal Article
In: 2025, (dagstat (2025)).
Explanation Regularisation through the Lens of Attributions Journal Article
In: 2024, (COLING (2025)).
SimuDICE: Offline Policy Optimization Through World Model Updates and DICE Estimation Journal Article
In: 2024, (BNAIC).
Use of sample-splitting and cross-fitting techniques to mitigate the risks of double-dipping in behaviour-agnostic reinforcement learning Journal Article
In: 2024, (BNAIC).
Understanding the Effects of the Baidu-ULTR Logging Policy on Two-Tower Models Journal Article
In: 2024, (CONSEQUENCES Workshop at RecSys (2024)).
Unbiased Learning to Rank Meets Reality: Lessons from Baidu's Large-Scale Search Dataset Journal Article
In: 2024, (SIGIR (2024)).
Leveraging diverse offline data in POMDPs with unobserved confounders Journal Article
In: 2024, (EWRL (2024)).
Dynamic Structural Causal Models Journal Article
In: 2024, (Causal Inference for Time Series Workshop @UAI (2024)).
Unbiased Learning to Rank: On Recent Advances and Practical Applications Journal Article
In: 2024, (WSDM (2024)).
Evaluating and Correcting Performative Effects of Decision Support Systems via Causal Domain Shift Journal Article
In: 2024, (CLeaR (2024)).
When Do Off-Policy and On-Policy Policy Gradient Methods Align? Journal Article
In: 2024, (arXiv).
Recent Advancements in Unbiased Learning to Rank Journal Article
In: 2024, (FIRE (2023)).
Report on the 1st Workshop on Generative Information Retrieval Workshop
2024, (ACM SIGIR Forum (2024)).
Collaborative filtering algorithms are prone to mainstream-taste bias Journal Article
In: 2023, (RecSys (2023)).
Correcting for Nonignorable Selection Bias and Missing Response in Regression using Privileged Information Journal Article
In: 2023, (UAI (2023)).
An Offline Metric for the Debiasedness of Click Models Journal Article
In: 2023, (SIGIR (2023)).
Recent Advances in the Foundations and Applications of Unbiased Learning to Rank Journal Article
In: 2023, (SIGIR (2023)).
Unbiased Neural Click Models and Pointwise IPS Rankers Journal Article
In: 2023, (ECIR (2023)).
Are Neural Click Models Pointwise IPS Rankers? Miscellaneous
2022, (CONSEQUENCES+REVEAL Workshop at RecSys (2022)).
Booking.com has grown from a small Dutch startup in 1996 to one of the world’s leading digital travel companies.
Delft University of Technology (TU Delft) is a technical university in Delft. Top education and research are at the heart of the oldest and largest technical university in the Netherlands.
University of Amsterdam (UvA) is the Netherlands’ largest university, offering the widest range of academic programmes.
