We use cookies to analyze the browsing and usage of our website and to personalize your experience. You can disable these technologies at any time, but this may limit certain functionalities of the site. Read our Privacy Policy for more information.
Setting cookies
You can enable and disable the types of cookies you wish to accept. However certain choices you make could affect the services offered on our sites (e.g. suggestions, personalised ads, etc.).
Essential cookies
These cookies are necessary for the operation of the site and cannot be deactivated. (Still active)
Analytics cookies
Do you accept the use of cookies to measure the audience of our sites?
Multimedia Player
Do you accept the use of cookies to display and allow you to watch the video content hosted by our partners (YouTube, etc.)?
Publications
Episodes Meta Sequence S 2 Fast Update Slow Update Fast Update Slow Update
Decomposing knowledge into interchangeable pieces promises a generalization advantage when there are changes in distribution. A learning age… (see more)nt interacting with its environment is likely to be faced with situations requiring novel combinations of existing pieces of knowledge. We hypothesize that such a decomposition of knowledge is particularly relevant for being able to generalize in a systematic manner to out-of-distribution changes. To study these ideas, we propose a particular training framework in which we assume that the pieces of knowledge an agent needs and its reward function are stationary and can be re-used across tasks. An attention mechanism dynamically selects which modules can be adapted to the current task, and the parameters of the selected modules are allowed to change quickly as the learner is confronted with variations in what it experiences, while the parameters of the attention mechanisms act as stable, slowly changing, metaparameters.We focus on pieces of knowledge captured by an ensemble of modules sparsely communicating with each other via a bottleneck of attention. We find that meta-learning the modular aspects of the proposed system greatly helps in achieving faster adaptation in a reinforcement learning setup involving navigation in a partially observed grid world with image-level input. We also find that reversing the role of parameters and meta-parameters does not work nearly as well, suggesting a particular role for fast adaptation of the dynamically selected modules.
Estimating the Impact of an Improvement to a Revenue Management System: An Airline Application
Airlines have been making use of highly complex Revenue Management Systems to maximize revenue for decades. Estimating the impact of changin… (see more)g one component of those systems on an important outcome such as revenue is crucial, yet very challenging. It is indeed the difference between the generated value and the value that would have been generated keeping business as usual, which is not observable. We provide a comprehensive overview of counterfactual prediction models and use them in an extensive computational study based on data from Air Canada to estimate such impact. We focus on predicting the counterfactual revenue and compare it to the observed revenue subject to the impact. Our microeconomic application and small expected treatment impact stand out from the usual synthetic control applications. We present accurate linear and deep-learning counterfactual prediction models which achieve respectively 1.1% and 1% of error and allow to estimate a simulated effect quite accurately.
Existing accounts of explanation emphasise 001 the role of prior experience and analogy in 002 the solution of new problems. However, most 0… (see more)03 of the contemporary models for multi-hop tex-004 tual inference construct explanations consider-005 ing each test case in isolation. This paradigm 006 is known to suffer from semantic drift, which 007 causes the construction of spurious explana-008 tions leading to wrong predictions. In con-009 trast, we propose an abductive framework for 010 multi-hop inference that adopts the retrieve - 011 reuse - revise paradigm largely studied in case-012 based reasoning . Specifically, we present 013 ETNA ( E xplana t io n by A nalogy), a novel 014 model that addresses unseen inference prob-015 lems by retrieving and adapting prior expla-016 nations from similar training examples. We 017 empirically evaluate the case-based abductive 018 framework on downstream commonsense and 019 scientific reasoning tasks. Our experiments 020 demonstrate that ETNA can be effectively in-021 tegrated with sparse and dense encoding mech-022 anisms or downstream transformers, achiev-023 ing strong performance when compared to ex-024 isting explainable approaches. Moreover, we 025 study the impact of the retrieve - reuse - revise 026 paradigm on explainability and semantic drift, 027 showing that it boosts the quality of the con-028 structed explanations, resulting in improved 029 downstream inference performance. 030
Exploring the Wasserstein metric for time-to-event analysis.
Survival analysis is a type of semi-supervised task where the target output (the survival time) is often right-censored. Utilizing this info… (see more)rmation is a challenge because it is not obvious how to correctly incorporate these censored examples into a model. We study how three categories of loss functions can take advantage of this information: partial likelihood methods, rank methods, and our own classification method based on a Wasserstein metric (WM) and the non-parametric Kaplan Meier (KM) estimate of the probability density to impute the labels of censored examples. The proposed method predicts the probability distribution of an event, letting us compute survival curves and expected times of survival that are easier to interpret than the rank. We also demonstrate that this approach directly optimizes the expected C-index which is the most common evaluation metric for survival models.
Factorizing Declarative and Procedural Knowledge in Structured, Dynamical Environments
Decomposing knowledge into interchangeable pieces promises a generalization advantage when there are changes in distribution. A learning age… (see more)nt interacting with its environment is likely to be faced with situations requiring novel combinations of existing pieces of knowledge. We hypothesize that such a decomposition of knowledge is particularly relevant for being able to generalize in a systematic manner to out-of-distribution changes. To study these ideas, we propose a particular training framework in which we assume that the pieces of knowledge an agent needs and its reward function are stationary and can be re-used across tasks. An attention mechanism dynamically selects which modules can be adapted to the current task, and the parameters of the selected modules are allowed to change quickly as the learner is confronted with variations in what it experiences, while the parameters of the attention mechanisms act as stable, slowly changing, meta-parameters. We focus on pieces of knowledge captured by an ensemble of modules sparsely communicating with each other via a bottleneck of attention. We find that meta-learning the modular aspects of the proposed system greatly helps in achieving faster adaptation in a reinforcement learning setup involving navigation in a partially observed grid world with image-level input. We also find that reversing the role of parameters and meta-parameters does not work nearly as well, suggesting a particular role for fast adaptation of the dynamically selected modules.
In this paper, we study the finite-time behaviour of temporal difference (TD) learning algorithms when combined with tail-averaging, and pr… (see more)esent instance dependent bounds on the parameter error of the tail-averaged TD iterate. Our error bounds hold in expectation as well as with high probability, exhibit a sharper rate of decay for the initial error (bias), and are comparable with existing bounds in the literature.
This paper is about the problem of learning a stochastic policy for generating an object (like a molecular graph) from a sequence of actions… (see more), such that the probability of generating an object is proportional to a given positive reward for that object. Whereas standard return maximization tends to converge to a single return-maximizing sequence, there are cases where we would like to sample a diverse set of high-return solutions. These arise, for example, in black-box function optimization when few rounds are possible, each with large batches of queries, where the batches should be diverse, e.g., in the design of new molecules. One can also see this as a problem of approximately converting an energy function to a generative distribution. While MCMC methods can achieve that, they are expensive and generally only perform local exploration. Instead, training a generative policy amortizes the cost of search during training and yields to fast generation. Using insights from Temporal Difference learning, we propose GFlowNet, based on a view of the generative process as a flow network, making it possible to handle the tricky case where different trajectories can yield the same final state, e.g., there are many ways to sequentially add atoms to generate some molecular graph. We cast the set of trajectories as a flow and convert the flow consistency equations into a learning objective, akin to the casting of the Bellman equations into Temporal Difference methods. We prove that any global minimum of the proposed objectives yields a policy which samples from the desired distribution, and demonstrate the improved performance and diversity of GFlowNet on a simple domain where there are many modes to the reward function, and on a molecule synthesis task.
