Fabian Bastin

Because household travel surveys combine household characteristics, variable-length person rosters, vehicle rosters, and trip chains, existing synthesis pipelines split them across separate models or flatten them into cropped fixed-length vectors. Representation, rather than modeling, becomes the bottleneck: splitting severs cross-segment dependence, and cropping truncates the household-structure distribution. We instead tokenize the diary and keep the object whole: each household-day becomes one token sequence with count tokens for vehicles, persons, and trips, and distinct symbols for padded, hidden, unobserved, and inapplicable positions. We introduce DiaryBridge, a masked diffusion model that bridges the fully hidden sequence to the distribution of complete diaries by unmasking positions in any order. One trained model, conditioned on any observed subset, performs synthesis and partial-diary completion alike. Experiments on real data extracted from the 2017/2018 Regional Travel Survey of the Baltimore region show that DiaryBridge generates travel reflecting vehicle availability and employment status, recovers shared trips using other household members’ observed diaries, and better reproduces aggregate demand across modes, purposes, and departure periods than the learned baselines.

Testing algorithmic trading strategies in production carries substantial risk of market disruption, yet passive historical replay cannot capture the market's reaction to an agent's actions. We propose a reactive limit order book (LOB) simulator based on a Conditional Variational Autoencoder (CVAE) coupled with an LSTM module that conditions generation on the ten most recent book snapshots. The model operates directly on state transitions signed, normalized depth vectors at three bid and ask levels rather than on heterogeneous order-by-order events, yielding a compact and homogeneous learning target. We train and evaluate on Toronto Stock Exchange (TSX) equity data, a market underrepresented in the simulation literature. The simulator reproduces key stylized facts: marginal queue distributions, the characteristic LOB smile in average quantities, level-to-level correlations, and realistic price paths-while exhibiting reactive price impact for both liquidity taking and provision, with the concave size-impact relationship qualitatively consistent with the literature. Benchmarked against a Conditional Wasserstein GAN with gradient penalty trained on the same data, the CVAE achieves superior queue-level fidelity and inter-level correlation structure with a simpler, single-network ELBO objective that requires neither gradient-penalty tuning nor critic-iteration scheduling, at the cost of a higher marginal KL divergence on the more liquid of the two stocks studied. To the best of our knowledge, this snapshot-conditioned CVAE+LSTM simulator is the first of its kind applied to Canadian equities, offering reactive synthetic environments for execution and market-making research.

The extended aircraft arrival management problem, as an extension of the classic aircraft landing problem, seeks to preschedule aircraft on a destination airport a few hours before their planned landing times. A two-stage stochastic mixed-integer programming model enriched by chance constraints is proposed in this paper. The first-stage optimization problem determines an aircraft sequence and target times over a reference point in the terminal area, called initial approach fix (IAF), so as to minimize the landing sequence length. Actual times over the IAF are assumed to deviate randomly from target times following known probability distributions. In the second stage, actual times over the IAF are assumed to be revealed, and landing times are to be determined in view of minimizing a time-deviation impact cost function. A Benders reformulation is proposed, and acceleration techniques to Benders decomposition are sketched. Extensive results on realistic instances from Paris Charles-de-Gaulle airport show the benefit of two-stage stochastic and chance-constrained programming over a deterministic policy.