AdONE Seminar - Monday, May 12, 2025: Prof. Flore Sentenac (HEC Paris)

Balancing optimism and pessimism in offline-to-online learning

We introduce the offline-to-online learning framework through an application to Click-Through-Rate (CTR) prediction in online advertising. A decision-maker starts with a fixed dataset of prior interactions collected without their control and uses it to iteratively refine their strategy, maximizing cumulative rewards. A key challenge arises: for short horizons, the pessimistic Lower Confidence Bound (LCB) algorithm performs well, competing with strategies supported by the offline data. For longer horizons, the optimistic Upper Confidence Bound (UCB) algorithm is preferred, as it converges to the optimal strategy at a near-optimal rate. However, UCB initially over-explores, leading to worse short-term performance than LCB. This creates a strategic dilemma: a decision-maker uncertain about the deployment duration should start with LCB and gradually transition to UCB as more interactions occur. We explore how and why this transition should happen. Our main result shows that our proposed algorithm performs nearly as well as the better of LCB and UCB at any point in time. 

Based on a paper with Ilbin Lee and Csaba Szeppesvari. 

Date: May 12, 2025

Time: 5 pm, s.t.

Place: Arcisstr. 21, room 0505.Z1.534Z

 

AdONE Seminar - Monday, May 19, 2025: Prof. Yaoxin Wu (Eindhoven University of Technology)

Towards unifying neural combinatorial optimization through large language model

To advance capabilities of large language models (LLMs) in solving combinatorial optimization problems (COPs), this paper presents the Language-based Neural COP Solver (LNCS), a novel framework that is unified for the end-to-end resolution of diverse text-attributed COPs. LNCS leverages LLMs to encode problem instances into a unified semantic space, and integrates their embeddings with a Transformer-based solution generator to produce high-quality solutions. By training the solution generator with conflict-free multi-task reinforcement learning, LNCS effectively enhances LLM performance in tackling COPs of varying types and sizes, achieving state-of-the-art results across diverse problems. Extensive experiments validate the effectiveness and generalizability of the LNCS, highlighting its potential as a unified and practical framework for real-world COP applications.

Date: May 19, 2025

Time: 5 pm, s.t.

Place: Arcisstr. 21, room 0505.Z1.534Z

 

AdONE Seminar - Monday, June 2, 2025: Prof. Roel Leus (KU Leuven)

A flow-based formulation for parallel machine scheduling using decision diagrams

We present a new flow-based formulation for identical parallel machine scheduling with a regular objective function and without idle time.  The formulation is constructed with the help of a decision diagram that represents all job sequences that respect specific ordering rules. These rules rely on a partition of the planning horizon into, generally non-uniform, periods and do not exclude all optimal solutions, but they constrain solutions to adhere to a canonical form.  The new formulation has numerous variables and constraints, and hence we apply a Dantzig-Wolfe decomposition in order to compute the linear programming relaxation in reasonable time; the resulting lower bound is stronger than the bound from the classical time-indexed formulation. We develop a branch-and-price framework that solves several instances from the literature for the first time. We compare the new formulation with the time-indexed and arc-time-indexed formulation by means of a series of computational experiments.

Date: June 2, 2025

Time: 5 pm, s.t.

Place: Arcisstr. 21, room 0505.Z1.534Z

 

AdONE Seminar - Monday, June 23, 2025: Prof. Daniel Kuhn (EPFL)

On the Interplay of Optimal Transport and Distributionally Robust Optimization

Optimal Transport (OT) seeks the most efficient way to morph one probability distribution into another one, and Distributionally Robust Optimization (DRO) studies worst-case risk minimization problems under distributional ambiguity. It is well known that OT gives rise to a rich class of data-driven DRO models, where the decision-maker plays a zero-sum game against nature who can adversely reshape the empirical distribution of the uncertain problem parameters within a prescribed transportation budget. Even though generic OT problems are computationally hard, the Nash strategies of the decision-maker and nature in OT-based DRO problems can often be computed efficiently. In this talk we will uncover deep connections between robustification and regularization, and we will disclose striking properties of nature's Nash strategy, which implicitly constructs an adversarial training dataset. We will also show that OT-based DRO offers a principled approach to deal with distribution shifts and heterogeneous data sources, and we will highlight new applications of OT-based DRO in machine learning, statistics, risk management and control. Finally, we will argue that, while OT is useful for DRO, ideas from DRO can also help us to solve challenging OT problems.

Date: June 23, 2025

Time: 5 pm, s.t.

Place: Boltzmannstr. 3, room 00.07.014

 

AdONE Seminar - Monday, June 30, 2025: Prof. Tobias Harks (Universität Passau)

Title: tba

Date: June 30, 2025

Time: 5 pm, s.t.

Place: Boltzmannstr. 3, room 00.04.011, MI Hörsaal 2 (5604.EG.011)

 

AdONE Seminar - Monday, July 14, 2025: Prof. Max Klimm (TU Berlin)

Title: tba

Date: July 14, 2025

Time: 5 pm, s.t.

Place: Boltzmannstr. 3, room tba