Horseshoe bats (Rhinolophidae) and the related Old World leaf-nosed bats (Hipposideridae) both show conspicuous pinna motions as part of their biosonar behaviors. In the current work, the kinematics of these motions in one species from each family (Rhinolophus ferrumequinum and Hipposideros armiger) has been analyzed quantitatively using three-dimensional tracking of landmarks placed on the pinna. The pinna motions that were observed in both species fell into two categories: In "rigid rotations" motions the geometry of the pinna was preserved and only its orientation in space was altered. In "open-close motions" the geometry of the pinna was changed which was evident in a change of the distances between the landmark points. A linear discriminant analysis showed that motions from both categories could be separated without any overlap in the analyzed data set. Hence, bats from both species have two separate types of pinna motions with apparently no transitions between them. The deformations associated with open-close pinna motions in Hipposideros armiger were found to be substantially larger compared to the wavelength associated with the largest pulse energy than in Rhinolophus ferrumequinum (137% vs 99%). The role of the two different motions in the biosonar behaviors of the animals remains to be determined.

Many bat species, e.g., in the rhinolophid and hipposiderid families, have dynamic biosonar systems with highly mobile pinnae. Pinna motion patterns have been shown to fall into two distinct categories: rigid rotations and non-rigid motions (i.e., deformations). In the present work, two questions regarding the rigid rotations have been investigated: (i) what is the nature of the variability (e.g., discrete subgroups or continuous variation) within the rigid motions, (ii) what is its acoustic impact? To investigate the first question, rigid pinna motions in Pratt's leaf-nosed bats (Hipposideros pratti) have been tracked with stereo vision and a dense set of landmark points on the pinna surface. Axis-angle representations of the recorded rigid motions have shown a continuous variation in the rotation axes that covered a range of almost 180° in azimuth and elevation. To investigate the second question, the observed range of rigid pinna motions has been reproduced with a biomimetic pinna. Normalized mutual information between acoustic inputs associated with every pair of the rigid pinna motions showed that even small changes in the rotation axis resulted in more than 50% new sensory information encoding capacity (i.e., normalized mutual information less than 50%). This demonstrates a potential sensory benefit to the observed variability in the rigid pinna rotations.

As part of their biosonar behaviors, rhinolophid (family Rhinolophidae) and hipposiderid bats (family Hipposideridae) both show conspicuous motions of their outer ears (pinna). These motions coincide with pulse emission and echo reception in time and could hence have a functional relevance for the encoding of sensory information. However, a quantitative in-depth characterization of these motions is still needed to derive detailed hypotheses for their function. To accomplish this, dense sets of landmark points have been placed on the pinna to provide for dense spatial coverage of its surface over the course of a motion cycle. Occlusion-free stereo tracking of the landmarks was accomplished with an array of four high-speed video cameras. Customized methods based on motion prediction have been used to track landmark points across video frames. The results have been used to construct accurate, continuous estimates of pinna surface motion. These estimates reveal that the pinna surface is subject to heterogeneous patterns of displacements and velocities within each motion cycle. Experiments with simplified robotic reproductions to understand the acoustic implications of these pinna surface motions are currently in progress. Once the signal transformations that result from the pinna motions are understood, the question of functional relevance can be addressed.

No abstract

This paper considers over-the-air federated learning (OTA-FL). OTA-FL exploits the superposition property of the wireless medium, and performs model aggregation over the air for free. Thus, it can greatly reduce the communication cost incurred in communicating model updates from the edge devices. In order to fully utilize this advantage while providing comparable learning performance to conventional federated learning that presumes model aggregation via noiseless channels, we consider the joint design of transmission scaling and the number of local iterations at each round, given the power constraint at each edge device. We first characterize the training error due to such channel noise in OTA-FL by establishing a fundamental lower bound for general functions with Lipschitz-continuous gradients. Then, by introducing an adaptive transceiver power scaling scheme, we propose an over-the-air federated learning algorithm with joint adaptive computation and power control (ACPC-OTA-FL). We provide the convergence analysis for ACPC-OTA-FL in training with non-convex objective functions and heterogeneous data. We show that the convergence rate of ACPC-OTA-FL matches that of FL with noise-free communications.