Submodular Observation Selection and Information Gathering for Quadratic Models

Authors: Abolfazl Hashemi, Mahsa Ghasemi, Haris Vikalo, Ufuk Topcu

ICML 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental To demonstrate efficacy of the proposed framework, we consider two applications multi-object tracking and phase retrieval and empirically verify that the subsets selected by the proposed greedy algorithm outperform approaches based on random selection, and greedy selection of observations that relies on a linearized model.
Researcher Affiliation Academia Abolfazl Hashemi 1 Mahsa Ghasemi 1 Haris Vikalo 1 Ufuk Topcu 1 1University of Texas at Austin.
Pseudocode Yes Algorithm 1 Greedy Observation Selection 1: Input: Utility function f(S), set of all observations X, number of selected observations k. 2: Output: Subset Sg X with |Sg| = k. 3: Initialize Sg = 4: for i = 0, . . . , k 1 do 5: js = argmaxj X\Sgfj(Sg) 6: Sg Sg {js} 7: end for
Open Source Code No No explicit statement or link for open source code is provided.
Open Datasets No We implement a Monte Carlo simulation with 50 independent instances where 10 moving objects are initially uniformly distributed in a 5 10 area. ... The UAVs can acquire range and angular measurements of the objects that are within the maximum radar detection range.
Dataset Splits No We implement a Monte Carlo simulation with 50 independent instances where 10 moving objects are initially uniformly distributed in a 5 10 area.
Hardware Specification No No specific hardware details are mentioned in the paper.
Software Dependencies No The paper mentions algorithms like Extended Kalman Filter and Wirtinger Flow, but does not provide specific software names with version numbers.
Experiment Setup Yes We implement a Monte Carlo simulation with 50 independent instances where 10 moving objects are initially uniformly distributed in a 5 10 area. At each time instance, the objects move in a random direction with a constant velocity set to 0.2. The swarm consists of 10 UAVs, equidistantly spread over the area... The maximum radar detection range is set such that, at each time step, the UAVs together collect approximately 130-170 range and angular measurements. The communication bandwidth constraints limit the number of measurements transmitted to the control unit to 10% of the gathered measurements. We consider two noise models... We use A-optimality and D-optimality as the selection criteria... complex signal θ Cn with n = 128 by selecting a subset of size k from m = 1280 observations. The complex signal θ is distributed as standard circularly-symmetric complex Gaussian, and we vary k from 256 to 576. We consider two standard scenarios where for each we average the results over 50 independent instances, and use the Wirtinger flow algorithm (Candes et al., 2015b) as the oracle estimator.