Alexander Ihler’s research while affiliated with University of California, Irvine and other places

What is this page?


This page lists works of an author who doesn't have a ResearchGate profile or hasn't added the works to their profile yet. It is automatically generated from public (personal) data to further our legitimate goal of comprehensive and accurate scientific recordkeeping. If you are this author and want this page removed, please let us know.

Publications (142)


A Deep Q-Learning based, Base-Station Connectivity-Aware, Decentralized Pheromone Mobility Model for Autonomous UAV Networks
  • Article

December 2024

IEEE Transactions on Aerospace and Electronic Systems

·

Alexander Ihler

·

Wireless networks consisting of low SWaP (size, weight, and power), fixed-wing UAVs (unmanned aerial vehicles) are used in many applications, such as search, monitoring and information gathering of inaccessible areas, in which UAVs sense within an area and forward the information, in a multihop manner, to an aerial base station (BS). Robustly performing these tasks requires the UAV network to be decentralized, autonomous, and scalable. An important tradeoff is between area coverage and connectivity: fast area coverage is needed to quickly identify objects of interest, while connectivity must be maintained for coordination and to transmit sensed information to the BS in real time. These factors must be balanced by the mobility model, which for each UAV has access only to locally available information. While [1], [2] attempt to balance these factors using flocking behavior, this only encourages the UAVs to spread, rather than using knowledge of what areas have already been covered. In this paper, we develop a neighborhood- and BS-connectivity aware distributed pheromone mobility model, called BS-CAP, to autonomously coordinate the UAV movements in a decentralized network. By using a pheromone map, we directly incorporate recent coverage information for the area. We then extend our approach to a deep Q-learning policy variant, called BSCAPDQN, to further tune and improve the balance between coverage and connectivity. These mobility models are fully distributed and rely only on information from neighboring UAVs. Our simulations demonstrate that both models achieve efficient area coverage and improved connectivity (both locally and to the BS), providing significant improvements over existing approaches.


Graph-based Complexity for Causal Effect by Empirical Plug-in

November 2024

This paper focuses on the computational complexity of computing empirical plug-in estimates for causal effect queries. Given a causal graph and observational data, any identifiable causal query can be estimated from an expression over the observed variables, called the estimand. The estimand can then be evaluated by plugging in probabilities computed empirically from data. In contrast to conventional wisdom, which assumes that high dimensional probabilistic functions will lead to exponential evaluation time of the estimand. We show that computation can be done efficiently, potentially in time linear in the data size, depending on the estimand's hypergraph. In particular, we show that both the treewidth and hypertree width of the estimand's structure bound the evaluation complexity of the plug-in estimands, analogous to their role in the complexity of probabilistic inference in graphical models. Often, the hypertree width provides a more effective bound, since the empirical distributions are sparse.


Estimating Causal Effects from Learned Causal Networks

October 2024

·

5 Reads

The standard approach to answering an identifiable causal-effect query (e.g., P(Y|do(X)) given a causal diagram and observational data is to first generate an estimand, or probabilistic expression over the observable variables, which is then evaluated using the observational data. In this paper, we propose an alternative paradigm for answering causal-effect queries over discrete observable variables. We propose to instead learn the causal Bayesian network and its confounding latent variables directly from the observational data. Then, efficient probabilistic graphical model (PGM) algorithms can be applied to the learned model to answer queries. Perhaps surprisingly, we show that this model completion learning approach can be more effective than estimand approaches, particularly for larger models in which the estimand expressions become computationally difficult. We illustrate our method’s potential using a benchmark collection of Bayesian networks and synthetically generated causal models.


Estimating Causal Effects from Learned Causal Networks

August 2024

·

4 Reads

·

1 Citation

The standard approach to answering an identifiable causal-effect query (e.g., P(Ydo(X)P(Y|do(X)) when given a causal diagram and observational data is to first generate an estimand, or probabilistic expression over the observable variables, which is then evaluated using the observational data. In this paper, we propose an alternative paradigm for answering causal-effect queries over discrete observable variables. We propose to instead learn the causal Bayesian network and its confounding latent variables directly from the observational data. Then, efficient probabilistic graphical model (PGM) algorithms can be applied to the learned model to answer queries. Perhaps surprisingly, we show that this \emph{model completion} learning approach can be more effective than estimand approaches, particularly for larger models in which the estimand expressions become computationally difficult. We illustrate our method's potential using a benchmark collection of Bayesian networks and synthetically generated causal models.


Fig. 1. Illustration of autonomous UAV ad-hoc network for remote monitoring of an inaccessible area, where a communication infrastructure is not available.
Fig. 2. Modules used in our H-AODV scheme.
Fig. 3. Pipe formation in our H-AODV scheme.
Fig. 5. Instantaneous PDR for H-AODV, LEPR and AODV schemes.
Fig. 6. Average PDR for H-AODV, LEPR and AODV schemes for different number of data flows and data rates, at varying node densities and speeds.

+1

A Hybrid Reactive Routing Protocol for Decentralized UAV Networks
  • Preprint
  • File available

July 2024

·

57 Reads

Wireless networks consisting of low SWaP, FW-UAVs are used in many applications, such as monitoring, search and surveillance of inaccessible areas. A decentralized and autonomous approach ensures robustness to failures; the UAVs explore and sense within the area and forward their information, in a multihop manner, to nearby aerial gateway nodes. However, the unpredictable nature of the events, relatively high speed of UAVs, and dynamic UAV trajectories cause the network topology to change significantly over time, resulting in frequent route breaks. A holistic routing approach is needed to support multiple traffic flows in these networks to provide mobility- and congestion-aware, high-quality routes when needed, with low control and computational overheads, using the information collected in a distributed manner. Existing routing schemes do not address all the mentioned issues. We present a hybrid reactive routing protocol for decentralized UAV networks. Our scheme searches routes on-demand, monitors a region around the selected route (the pipe), and proactively switches to an alternative route before the current route's quality degrades below a threshold. We empirically evaluate the impact of pipe width and node density on our ability to find alternate high-quality routes within the pipe and the overhead required to maintain the pipe. Compared to existing reactive routing schemes, our approach achieves higher throughput and reduces the number of route discoveries, overhead, and resulting flow interruptions at different traffic loads, node densities and speeds. Despite having limited network topology information, and low overhead and route computation complexity, our proposed scheme achieves superior throughput to proactive optimized link state routing scheme at different network and traffic settings. We also evaluate the relative performance of reactive and proactive routing schemes.

Download

Boosting AND/OR-Based Computational Protein Design: Dynamic Heuristics and Generalizable UFO

August 2023

·

3 Reads

Scientific computing has experienced a surge empowered by advancements in technologies such as neural networks. However, certain important tasks are less amenable to these technologies, benefiting from innovations to traditional inference schemes. One such task is protein re-design. Recently a new re-design algorithm, AOBB-K*, was introduced and was competitive with state-of-the-art BBK* on small protein re-design problems. However, AOBB-K* did not scale well. In this work we focus on scaling up AOBB-K* and introduce three new versions: AOBB-K*-b (boosted), AOBB-K*-DH (with dynamic heuristics), and AOBB-K*-UFO (with underflow optimization) that significantly enhance scalability.


Design Amortization for Bayesian Optimal Experimental Design

June 2023

·

8 Reads

·

4 Citations

Proceedings of the AAAI Conference on Artificial Intelligence

Bayesian optimal experimental design is a sub-field of statistics focused on developing methods to make efficient use of experimental resources. Any potential design is evaluated in terms of a utility function, such as the (theoretically well-justified) expected information gain (EIG); unfortunately however, under most circumstances the EIG is intractable to evaluate. In this work we build off of successful variational approaches, which optimize a parameterized variational model with respect to bounds on the EIG. Past work focused on learning a new variational model from scratch for each new design considered. Here we present a novel neural architecture that allows experimenters to optimize a single variational model that can estimate the EIG for potentially infinitely many designs. To further improve computational efficiency, we also propose to train the variational model on a significantly cheaper-to-evaluate lower bound, and show empirically that the resulting model provides an excellent guide for more accurate, but expensive to evaluate bounds on the EIG. We demonstrate the effectiveness of our technique on generalized linear models, a class of statistical models that is widely used in the analysis of controlled experiments. Experiments show that our method is able to greatly improve accuracy over existing approximation strategies, and achieve these results with far better sample efficiency.




Fig. 1: Illustration of next-waypoint selection based on repel pheromone intensity and connectivity of the UAV.
Fig. 2: (a) UAV heading discretized into 8 directions, (b) Selection of next-waypoint cells for a fixed-wing UAV.
Fig. 3: Performance Curves for 20 and 40 UAVs at 20 m/s.
Connectivity-Aware Pheromone Mobility Model for Autonomous UAV Networks

October 2022

·

54 Reads

UAV networks consisting of reduced size, weight, and power (low SWaP) fixed-wing UAVs are used for civilian and military applications such as search and rescue, surveillance, and tracking. To carry out these operations efficiently, there is a need to develop scalable, decentralized autonomous UAV network architectures with high network connectivity. However, the area coverage and the network connectivity requirements exhibit a fundamental trade-off. In this paper, a connectivity-aware pheromone mobility (CAP) model is designed for search and rescue operations, which is capable of maintaining connectivity among UAVs in the network. We use stigmergy-based digital pheromone maps along with distance-based local connectivity information to autonomously coordinate the UAV movements, in order to improve its map coverage efficiency while maintaining high network connectivity.


Citations (75)


... To this end, recent work has replaced the simpler density approximations of Foster et al. [25] with more flexible density approximations based on transportation of measure. For instance, Orozco et al. [47], Dong et al. [22], Kennamer et al. [33] use conditional normalizing flows [55,49] trained from samples of π X,Y to approximate the density of π X|Y . Koval et al. [35] instead construct functional tensor-train approximations [18] of the triangular Knothe-Rosenblatt (KR) transport map using direct evaluations of the joint density π X,Y . ...

Reference:

Expected Information Gain Estimation via Density Approximations: Sample Allocation and Dimension Reduction
Design Amortization for Bayesian Optimal Experimental Design
  • Citing Article
  • June 2023

Proceedings of the AAAI Conference on Artificial Intelligence

... the Reference Point Group Mobility Model (RPGM) [27] is a widely used model for traditional MANETs, but very few group models are specifically developed for FANETs. A pheromone model was proposed in [28] for cooperative ad hoc networks of UAVs, but its pheromone logic can push UAVs away from each other, leading to the break of node links, making it not network-friendly. The existing network issues arise due to the fluctuating topology and high mobility, which increase networking problems. ...

A Deep Q-Learning Connectivity-Aware Pheromone Mobility Model for Autonomous UAV Networks
  • Citing Conference Paper
  • February 2023

... However, this approach is unsuitable for our GNSS-denied scenario because it relies on UAVs having knowledge of a predefined grid and requires communicating positional information about visited cells. Finally, a connectivity-aware pheromone-based model for UAV networks was proposed, requiring UAVs to communicate their positions to maintain network connectivity [34]. While this method is useful for maintaining connectivity, it is not feasible for coverage tasks in GNSS-denied environments, as it relies on position sharing among UAVs. ...

A Connectivity-Aware Pheromone Mobility Model for Autonomous UAV Networks
  • Citing Conference Paper
  • January 2023

... Garg et al. [22] developed a cross-layer, mobility, and congestion-aware routing protocol for UAV networks. Their work is quite an indication of the emerging need to take in more abstract network layers concurrently, and two to consider dynamic aspects like mobility and congestion in the UAV communication system. ...

A Cross-Layer, Mobility and Congestion-Aware Routing Protocol for UAV Networks

IEEE Transactions on Aerospace and Electronic Systems

... The subgradient method is different from the descent method [30], and there is usually no dominating stopping criterion. In real applications, we always keep track of the best solution, which is expressed as g ...

Fast Parallel and Adaptive Updates for Dual-Decomposition Solvers
  • Citing Article
  • August 2011

Proceedings of the AAAI Conference on Artificial Intelligence

... On the other hand, the depth-first scheme can generate suboptimal solutions faster. Hybrids of both schemes can produce upper and lower bounds from each component and empirical evaluation demonstrated the effectiveness of best+depth-first search algorithms (Marinescu et al. 2017). The AND/OR search space for IDs can be defined by incorporating expected utilities as the value of each node. ...

Anytime Best+Depth-First Search for Bounding Marginal MAP

Proceedings of the AAAI Conference on Artificial Intelligence

... In future work, ADAPART may be used to estimate general partition functions if a general upper bound [50,34,35] is found to nest with few calls to Refine. The matrix permanent specific implementation of ADAPART may benefit from tighter upper bounds on the permanent. ...

Anytime Anyspace AND/OR Search for Bounding the Partition Function
  • Citing Article
  • February 2017

Proceedings of the AAAI Conference on Artificial Intelligence

... Designated ORACLEs 6 [56], they share a lot of synergy with brokers and marshals and are an important component of the overall software infrastructure needed to support survey science. Such systems are similar to those for selecting objects to obtain spectra with the goal of overcoming Malmquist bias in training data [49,33,5] except that the prioritization is based on science utility. ...

Active learning with RESSPECT: Resource allocation for extragalactic astronomical transients
  • Citing Conference Paper
  • December 2020

Noble Kennamer

·

·

Santiago Gonzalez-Gaitan

·

[...]

·

... The algorithm takes a CPD graphical model as input (Definition 2.2) and outputs the K * MAP value of an optimal amino acid assignment to the residues. The algorithm is based on a class of AND/OR search algorithms over graphical models for optimization and inference tasks [Marinescu et al., 2014] and is empowered by constraint propagation. ...

AND/OR Search for Marginal MAP
  • Citing Article
  • January 2018

Journal of Artificial Intelligence Research

... The best-first component aims at generating improved upper bounds while the depth-first search facilitates the generation of improved lower bounds, which are obtained by solving exactly the corresponding conditioned summation subproblems. For this reason, these methods are limited to problems with tractable summation subproblems (Marinescu et al. 2017). ...

Anytime Best+Depth-First Search for Bounding Marginal MAP
  • Citing Conference Paper
  • December 2020