Jad Soucar

Jad Soucar

Ph.D Student @ University of Southern California

Talks and presentations

Blind Multi-Stage Scoring Auctions with Value Function Estimation

November 01, 2025

Talk, International Conference on Industrial Logistics, Tokyo, Japan

Government procurement often suffers from subjectivity because reviewers rely on variable and partially unknown value functions to score multi-attribute bids. We address this challenge with a few-shot value function estimation framework and propose a greedy online heuristic solver that incrementally approximates both the structure and weights of the underlying value function. We apply the algorithm to a public works case study, where our results show that a 7-attribute value function can be accurately recovered in under 20 iterations, enabling bidders to adapt quickly and effectively in multi-stage scoring auctions.

Download Slides

A Pruned Shortest Path Algorithm for L_0 Optimization Problems with Rockafellian Perturbations

October 24, 2025

Poster, Capital-One CREDIF Symposium, Los Angeles, CA, USA

In this presentation we propose an optimization-based approach to timeseries breakpoint detection and attribution, where the target time-series is a function of an arbitrary number of “feature” timeseries. We tackle this problem by first introducing a general loss function with $L_0$ regularization along with a dual formulation. We then reformulate the dual into a shortest path problem which we solve by developing several specialized lazy path finding algorithms. We conclude with some theoretical analysis of the efficiency of our proposed algorithms along with a use case in marketing strategy

Download Slides

A Quantum Statistical Model of Decision Making in a Single-Cell Eukaryote

July 30, 2025

Poster, 47th Annual Meeting of the Cognitive Science Society, San Francisco, CA, USA

The single-celled protist Stentor roselii has long been observed to exhibit complex decision-making behaviors, yet existing machine learning and classical computational models have struggled to replicate its actions. In this paper, we propose a novel quantum-statistical framework to model S. roselii’s behavioral responses to environmental stimuli. By leveraging quantum circuits with amplitude dampening and memory effects, we construct a quantum behavioral model that captures the probabilistic and hierarchical nature of S. roselii’s decision making. Our results suggest that quantum statistical theory provides a promising tool for representing and simulating biological decision processes.

Download Slides

The Ontology and Architecture of Intelligence

July 30, 2025

Talk, Social Intelligence Workshop @ Peking University , Beijing, China

We suggest the problem of intelligence is to locate viable solutions in infinite search spaces using finite resources. Here we present our work on conceptualizing and formalizing some of the features of intelligence in dynamic multi-level control systems. We begin by presenting a new ontology needed for understanding intelligence, illustrated by the controversy around beables in quantum theory. We then present a model of situationally appropriate information-processing in a single-celled organism, using quantum computing formalisms. We show how multi-level control systems can emerge from information-processing constraints within a system of multi-dimensional optimization and explore the information flow between levels within single cells. We then model how bottom-up and top-down prospection gives rise to flexible goals and strategies in wayfinding control systems, using examples from navigation and manual crafts. Finally, we model how prospection overshoot and undershoot can lead to run-away failures and how emotions can play a role in anticipating bad outcomes and dampen the oscillations, using examples from advertising and economics.

Download Slides

Classical Simulation of Non-Clifford Noise Channels

January 10, 2025

Talk, Join Mathematical Meeting, Seattle, WA, USA

We introduce the basics of quantum computing and simulation of quantum systems on classical computers. We then discuss noise in quantum systems and how it is classically modelled, along with the difficulties of simulating quantum noise on classical computers. Our primary question is how to efficiently simulate quantum noise by leveraging existing techniques based upon the Gottesman-Knill theorem, which provides efficient simulation of circuits containing only Clifford gates. To this end we develop theory and practical implementations of the K-Gadget, a method to compress and simualte thousands of instances of dampening noise within reasonable memory constraints.

Download Slides

Modeling Trust Based Markets

July 26, 2024

Talk, 46th Annual Meeting of the Cognitive Science Society, Rotterdamn, Netherlands

Cognitive processes underlie economic relations. In this talk, we discuss 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. Customers conversely orient towards advertisers seeking information or turn away from them as unreliable communicators. These behaviors and the patterns they generate occur inside a state space of unallocated perceived value. They constitute a small subset of the full range of possible strategic and adaptive responses that define cognitive microeconomics.

Download Slides