Jad Soucar

Jad Soucar

Ph.D Student @ University of Southern California

Research

Psuedo-Data Injections for CLO Bandit Problems

2026

Contextual linear optimization (CLO) with bandit feedback is a class of CLO problems where only the costs of historical actions are observable. Finding an optimal decision making policy in this setting suffers from the fundamental challenge that real-world data often lacks coverage over the action space, making the full cost vector unidentifiable with the data available. A common remedy is to apply regularization to ensures stability of the learning problem. We show that this approach admits an alternative interpretation as a specific form of pseudo-data injection where synthetic data is added to induce coverage. This perspective suggests a broader question...

Download Paper

Interpretable State and Time Dependent Multi-Touch Attribution

2026

Multi-touch attribution (MTA) aims to assign credit to the sequence of ads that influence a customer’s decision to make a purchase. Existing state-of-the-art models often rely on complex black-box predictors with post-hoc attribution (e.g., Shapley values), which can be unstable and difficult for industry to act on. We propose an interpretable, state and time-dependent MTA framework that explicitly models how advertising exposures accumulate and decay in a customer’s latent willingness to purchase. When coupled across customers the resulting problem is formulated as a mixed-integer problem, which we tackle by proposing the application of a family of scalable ADMM and quadratic-penalty...

Download Paper

Silos and Lazy Shortest Paths on Ordered Directed Acyclic Graphs

2025

Many dynamic programs can be interpreted as shortest path problems on ordered directed acyclic graphs (DAGs), where edge weights are optimal values of non-trivial optimization problems. In such cases using approximate lower-bounding weights can reduce computational cost. In this paper we introduce general formalisms to study these lazy shortest path problems where edge weight computation is delayed or avoided. Our primary contribution in this area is introducing the concept of a graph Silo, which captures the degree to which a graph permits paths that are nearly tied to the shortest path. We show that such formalisms are especially useful in...

Download Paper

Blind Multi-Stage Scoring Auctions with Two-Sided Uncertainty

2025

In this paper, we analyze multi-round scoring auctions where the auctioneers value function is unknown. We develop a greedy algorithm capable of multi-attribute value function estimation using information from only a few rounds of bidding. We apply our analysis to the case study of public works procurment.

Download Paper

Simulating Quantum Circuits with Non-Clifford Noise

2024

We introduce the basics of quantum computing and simulation of quantum systems on classical computers. We then discuss noise in quantum systems and how instances of noise are classically modelled, along with the difficulties of simulating quantum noise on classical computers. We introduce an extension of the T-Gadget to classically simulate thousands of instances of dampening noise within reasonable memory constraints.

Download Paper

Autonomous Trading Using Deep Q Learning

2024

In this paper, we explore the application of Deep Reinforcement Learning (DRL) to the domain of autonomous equity trading, with a particular focus on the use of Deep Q Networks (DQNs) coupled with risk-sensitive loss objectives, to develop trading agents capable of navigating complex financial market conditions.

Download Paper

A Model for Trust Driven Advertising

2024

In this paper, we develop a conceptual, mathematical, and computational framework for modeling market exchange as a series of dynamically interacting cognitive processes. Specifically, we show how advertisers can build trust and gain confidence in their pricing power to the point that they erode trust and undermine the efficacy of their advertising.

Download Paper