There are numerous examples of biological systems outperforming human-designed algorithms at engineering tasks. Rather than relying on one central controller, these systems rely on distributed agents following simple local interaction rules that yield complex collective behavior. My research studies these biological systems systems in order to reverse-engineer optimization algorithms, with a focus on designing robust, efficient routing networks.

Repairing broken links is a fundamental problem for human-designed routing networks, as well as trail networks used by ants to travel between nests and food sources. Many ant species have evolved methods to solve the repair problem under more restrictive constraints than those faced by human engineers. Arboreal turtle ants are especially interesting, because their movements are constrained by the branches in the vegetation. I defined a mathematical model for the unique repair problem faced by turtle ants. I demonstrated techniques for using turtle ant behavioral data to derive a distributed repair algorithm and infer critical features of turtle ant behavior. The algorithm is biologically realistic, scalable, and robust to a variety of topologies. I then analyzed data on trail networks formed by turtle ants over several days to conclude that turtle ant trail networks eschew shorter branches in favor of nodes whose physical orientations allow for quicker pheromone reinforcement.

Human-designed routing networks often seek to maximize efficiency while minimizing cost. These objectives often conflict, which necessitates manage trade-offs. Neural arbors face a similar challenge of conducting signals quickly while conserving material. I defined a mathematical framework for the multi-objective optimization problem that neural arbors face. I showed that neural arbors are significantly better at optimizing trade-offs between material cost and signal delay than would be expected by chance. I showed how this framework can be used to predict an arbor's structure based on its function. My work confirms the value of using neural arbors as inspiration for network design algorithms, and identifies promising new avenues for neuroanatomy research.

My research highlights the symbiotic relationships between biology and computer science: collecting and analyzing behavioral data can inform algorithm design, while algorithmic thinking can elucidating novel behavioral patterns.

Despite considerable effort over the past several decades, it is unclear how an organism’s nervous system extracts ecologically relevant information about its environment. The connections between individual neurons (i.e. the structural connectome), and the ways in which these connections are used (i.e. the functional connectome), likely constrain how a nervous system performs this feat. The nematode Caenorhabditis elegans (C. elegans) presents a rare opportunity to study both of these aspects of a nervous system, since the complete structural connectome of one individual has been published, and recent technological advances enable scientists to record from a large part of its nervous system. I studied the C. elegans nervous system from both of these perspectives to better understand the structural features and network-level dynamics that allow it to extract information about the environment.

I found that the C. elegans structural connectome, along with those of several other biological networks, display Rentian scaling – a power-law relationship between the number of nodes in a module and the number of connections to nodes in other modules. This indicates that these biological networks, but not other social and technological networks, must negotiate a trade-off between the efficiency of information transfer and cost to maintain connections. Thus, Rentian scaling may be a feature unique to information-processing networks, either because it presents an organizational principle used to process information, or because it indicates the existence of fundamental design constraints these networks share.

While a structural connectome constrains which neurons may communicate, a functional connectome can indicate which pathways of information transfer are actually used, with the caveat that statistical dependencies may reveal interactions that are actually false positives. I recorded from most neurons in the brain of adult C. elegans, and characterized their inferred functional connectome using tools from graph theory. I found that some patterns of activity were consistently modulated in a valence- or identity-specific manner. Moreover, these patterns could be used to identify which chemical an animal was experiencing. This indicates that studying the functional connectome of large-scale neural recordings may provide a mechanistic account of how neural networks extract and represent information about their environment.

Developing automated methods to efficiently process large volumes of point cloud data remains a grand challenge for 3D plant phenotyping applications. Phenotyping is one of the key pre-cursors towards developing more efficient breeding strategies and improvement of crop yields. Using 3D point clouds for plant phenotyping helps to eliminate problems that are usually encountered by performing phenotyping with images. However, point clouds come with their own set of problems, which include registration errors, missing data, sensor noise, and scalability. Point clouds of plants also typically have problems related to ``natural'' noise caused by small features such as trichomes or natural shape irregularities; further, handling the diversity of organic structures observed across plant species is challenging. All these nuances need to be considered to ensure quality extraction of phenotyping traits.

Here, we develop a collection of graph-theoretic and machine learning methods to tackle three primary challenges in plant phenotyping: leaf/branch classification, leaf counting, and branch skeletonization. For classification, we assess and validate the accuracy of our methods on a dataset of 54 3D shoot architectures representing multiple species, growth conditions, and developmental time-points. Using deep learning, we achieved 97.8% accuracy on classifying leaves versus branches; critically, we also demonstrate the robustness of our method to growth conditions and species that have not been trained on, which is important in practical applications but is often untested. For leaf counting, we developed an enhanced region growing algorithm to reduce over-segmentation; this method achieved 86.6% accuracy, outperforming prior methods developed for this problem. Finally, for branch skeletonization, we developed an enhanced tip detection technique that ran an order of magnitude faster and generated more precise skeleton architectures compared to prior methods. In addition, we improved the quality of angles produced by skeletonization. Overall, our improvements enable higher throughput and accurate extraction of phenotypic properties from 3D point cloud data. Finally, all the algorithmic methods presented in this thesis were packaged in a stand alone GUI application, called Plant3D (P3D).