Search for: All records

Several species of Polistes paper wasp are well known for their social hierarchies and the ability for individual wasps to modulate their social behaviors based on recognizable facial features of other wasps. For example, wasps that observe an aggressive social interaction between two other wasps will later behave differently toward the winner and loser of that interaction. Being able to alter the physical appearance of wasps~(e.g., with paint) has allowed for testing hypothetical roles of individual recognition in hierarchy formation, which is how researchers know that wasps are attending to faces specifically. However, these physical methods are limited in their scope. Social insects who respond to visual stimuli from other insects have been shown to give the same responses to playbacks of video recordings of those stimuli, which suggests that there may be a role for generative methods in social-insect research. Being able to computationally change the faces of individual wasps in a video recording of wasp social interactions would greatly expand the experimental toolbox of the behavioral researcher. Toward this end, we evaluate the use of an existing annotation-free model for image animation by motion transfer, the thin-plate spline motion model, for creating realistic videos that depict the face of a paper wasp performing behaviors recorded by another. Not needing to pre-define important landmarks is a strength of this method for this application space, but we find that "deep faking wasps" poses unique and non-trivial problems that still need to be solved before off-the-shelf motion transfer models can be used in the insect behavioral laboratory. 

Geometric Sensitive Hashing functions, a family of Local Sensitive Hashing functions, are neural network models that learn class-specific manifold geometry in supervised learning. However, given a set of supervised learning tasks, understanding the manifold geometries that can represent each task and the kinds of relationships between the tasks based on them has received little attention. We explore a formalization of this question by considering a generative process where each task is associated with a high-dimensional manifold, which can be done in brain-like models with neuromodulatory systems. Following this formulation, we define Task-specific Geometric Sensitive Hashing and show that a randomly weighted neural network with a neuromodulation system can realize this function. 

Many interpretable AI approaches have been proposed to provide plausible explanations for a model’s decision-making. However, configuring an explainable model that effectively communicates among computational modules has received less attention. A recently proposed shared global workspace theory showed that networks of distributed modules can benefit from sharing information with a bottle-necked memory because the communication constraints encourage specialization, compositionality, and synchronization among the modules. Inspired by this, we propose Concept-Centric Transformers, a simple yet effective configuration of the shared global workspace for interpretability, consisting of: i) an object-centric-based memory module for extracting semantic concepts from input features, ii) a cross-attention mechanism between the learned concept and input embeddings, and iii) standard classification and explanation losses to allow human analysts to directly assess an explanation for the model’s classification reasoning. We test our approach against other existing concept-based methods on classification tasks for various datasets, including CIFAR100, CUB-200-2011, and ImageNet, and we show that our model achieves better classification accuracy than all baselines across all problems but also generates more consistent concept-based explanations of classification output. 

Alarm signal propagation through ant colonies provides an empirically tractable context for analysing information flow through a natural system, with useful insights for network dynamics in other social animals. Here, we develop a methodological approach to track alarm spread within a group of harvester ants, Pogonomyrmex californicus . We initially alarmed three ants and tracked subsequent signal transmission through the colony. Because there was no actual standing threat, the false alarm allowed us to assess amplification and adaptive damping of the collective alarm response. We trained a random forest regression model to quantify alarm behaviour of individual workers from multiple movement features. Our approach translates subjective categorical alarm scores into a reliable, continuous variable. We combined these assessments with automatically tracked proximity data to construct an alarm propagation network. This method enables analyses of spatio-temporal patterns in alarm signal propagation in a group of ants and provides an opportunity to integrate individual and collective alarm response. Using this system, alarm propagation can be manipulated and assessed to ask and answer a wide range of questions related to information and misinformation flow in social networks.