Events

Moving goods in a globalized world is not easy but with automation to one integrated platform for full visibility and see better performance is possible. Currently, we leverage on technology to automate our shipment-based tracking on delivery status, respond to immediate adjustments and provide data to our clients. We foresee the future with shift to using robotic forklifts, automated storage systems in warehouses with robots picking and retrieving the goods on the shelves. Manpower equipped with specialized skills is crucial for adoption of these new technologies.
We must still create an identity through experience, commitment and working collaboratively are important; we want to ingrain our mission statement to all staff “People before Profit – Getting it Right First Time, Every Time”.

Motivated by emerging energy-intensive applications in electric vehicle charging and cloud computing, my research has focused on large-scale stochastic deadline scheduling problems under random task arrivals and processing cost. The objective is to minimize the expected sum of stochastic processing cost (due to intermittent renewable generation and fluctuating energy prices) and delay penalty cost (resulting from failures to finish tasks before user-specified deadlines). The hardness in these problems stems from the unknown dynamics of system uncertainties, and the high dimensionality in both the system state and action spaces. This talk overviews our research results on the rigorous establishment of structural optimal policy characterizations under discrete/continuous action spaces and convex/discrete delay penalties. The established optimal policy characterizations do not rely on any probabilistic assumption on the evolution of system uncertainties, and therefore can be naturally integrated into data-driven deep reinforcement learning (DRL) approaches for action space dimensionality reduction without loss of optimality. Numerical results on real-world data show that the proposed approach outperforms state-of-the-art DRL and MPC (model predictive control) based approaches.

Clustering is a fundamental problem in unsupervised learning. In this talk, we will introduce a convex clustering model, which can be regarded as a convex relaxation to the K-means with some favorable properties. We will establish sufficient conditions for the perfect recovery guarantee of the general weighted convex clustering model, which also improves existing theoretical results of the convex clustering model with uniform weights. In addition, we will introduce a highly efficient algorithm and a dimension reduction technique for solving the convex clustering model. Extensive numerical results will also demonstrate the superior performance of the convex clustering model and the proposed algorithms.

In this talk, we will present a framework on prescriptive analytics that has been successfully applied to several clients’ projects including a major US airline and a large financial institution. The framework consists of a causal teacher model which produces counterfactual outcomes corresponding to different treatment actions, and a prescriptive student model which distills a set of policies that optimizes a given objective in the form of a tree. The prescriptive tree can be built greedily as demonstrated in our ICML 2021 paper. As the greedy heuristic is unable to incorporate constraints that are ubiquitous, in our AAAI 2022 publication, we introduce a scalable mixed-integer program (MIP) approach that solves the constrained prescriptive policy generation problem via column generation.

Social science and humanities research could help overcome these weaknesses. The chapter shows how South Asian culture influences engineering practice and offers suggestions for social science and humanities scholars seeking rewarding research opportunities.

Mingmei Li & Benjamin Tan (Singapore University of Technology and Design) – Presentations by the Aviation Studies Institute

Very often, modern datasets in operations research have a panel structure — we observe data for many of distinct “units”, but for each unit we observe only a handful of relevant data points. As an example, consider a large online retailer where we observe data from thousands of distinct products, but each product typically has only a few sales. The dominant intuition when solving stochastic optimization problems in such settings is that we should “learn from similar units”, e.g., we might use covariates to cluster similar units and pool their data together when solving optimization problems. This intuition (in some form or the other) pervades most modern approaches to contextual stochastic optimization. Conversely, this intuition also suggests that if units are not “similar” in any way, aggregating data can only hurt us and we might as well treat each unit separately.

ESD Fall Newsletter 2022

What is the minimum car fleet size of a ride-hailing platform required to serve a given transport demand? In a sharing economy do prices yield unique results? Do crowdsourced platforms operate as efficiently as the ones where operations are managed centrally?

In this talk, we focus on two estimation problems: i) sparse regression and ii) sparse and smooth signal recovery. The first one is known to be NP-hard; we show that the second one is equivalent to a submodular minimization problem and, hence, it is polynomially solvable. For both problems, we derive a sequence of strong conic relaxations. The proposed rank-one strengthening can be interpreted as a non-separable, non-convex, unbiased sparsity-inducing regularizer, which dynamically adjusts its penalty according to the shape of the estimation error function without inducing bias for the sparse solutions. Computational experiments with benchmark datasets show that the proposed conic formulations are solved fast and result in near-optimal estimators for non-convex L0-problems. Moreover, the resulting estimators also outperform L1 approaches from a statistical perspective, achieving high prediction accuracy and good interpretability.