absorb.md

Ze Frank

Chronological feed of everything captured from Ze Frank.

Network Externalities and Component-Efficient Allocation

The paper introduces the Balanced Component-Efficient (BCE) rule for allocating contributions in networks with externalities, where the value of a component depends on the entire network structure. This rule ensures balanced contributions across all network edges, even though its construction primarily uses spanning-tree edges. A novel cycle-sum identity facilitates the generalization of balanced contributions from spanning trees to non-tree edges.

Iterative Problem Solving for TSPTW with Changing Time Window Constraints Outperforms From-Scratch Methods

This paper investigates the effectiveness of an iterative problem-solving protocol for the Traveling Salesperson Problem with Time Windows (TSPTW) when constraints change across related tasks. The protocol initializes each task using the best tour from the previous task, contrasting with a standard from-scratch approach. The key finding is that the iterative method consistently performs better, particularly in progressive-relaxation scenarios and on more challenging instances, demonstrating its potential for optimizing solutions to sequences of similar TSPTW problems.

Multitasking Pareto Optimization for Submodular Functions with Shared Constraints

This paper introduces multitasking formulations for solving multiple related monotone submodular problems with knapsack constraints. By sharing a common submodular function but having different constraints, the proposed approach allows for efficient solution sharing between problems, leading to significant performance improvements over independent classical methods. The research includes rigorous runtime analysis and experimental validation using the maximum coverage problem.

PIE-V: A Framework for Generating and Benchmarking Egocentric Procedural Videos with Mistakes

The PIE-V framework addresses the scarcity of egocentric procedural video data containing naturalistic human errors and subsequent recoveries. It leverages psychology-informed error and correction planning, LLMs for content generation and validation, and video synthesis to create realistic mistake-aware video sequences. This enables robust benchmarking of mistake detection and correction systems in egocentric computer vision.