Kotaro Ito | Searcharxiv

arXiv2023

Influence of self-disassembly of bridges on collective flow characteristics of swarm robots in a single-lane and periodic system with a gap

Inspired by the living bridges formed by ants, swarm robots have been developed to self-assemble bridges to span gaps and self-disassemble them. Self-disassembly of bridges (SDB) may increase the transport efficiency of swarm robots by increasing the number of moving robots, and also may decrease the efficiency by causing gaps to reappear. Our aim is to elucidate the influence of SDB on the collective flow characteristics of swarm robots in a single-lane and periodic system with a gap. In the system, robots span and cross the gap by self-assembling a single-layer bridge. We consider two scenarios in which SDB is prevented (prevent-scenario) or allowed (allow-scenario). We represent the horizontal movement of robots with a typical car-following model, and simply model the actions of robots for self-assembling and self-disassembling bridges. Numerical simulations have revealed the following results. Flow-density diagrams in both the scenarios shift to the higher-density region as the gap length increases. When density is low, allow-scenario exhibits the steady state of repeated self-assembly and self-disassembly of bridges. If density is extremely low, flow in this state is greater than flow in prevent-scenario owing to the increase in the number of robots moving horizontally. Otherwise, flow in this state is smaller than flow in prevent-scenario. Besides, flow in this state increases monotonically with respect to the velocity of robots in joining and leaving bridges. Thus, SDB is recommended for only extremely low-density conditions in periodic systems. Moreover, we have found hysteresis under the absence of periodic boundary conditions (in an open system). This study contributes to the development of the collective dynamics of self-driven particles that self-assemble structures, and stirs the dynamics with other self-assembled structures, such as ramps, chains, and towers.

physics.soc-ph↗

arXiv2020

Spectral Pruning: Compressing Deep Neural Networks via Spectral Analysis and its Generalization Error

Compression techniques for deep neural network models are becoming very important for the efficient execution of high-performance deep learning systems on edge-computing devices. The concept of model compression is also important for analyzing the generalization error of deep learning, known as the compression-based error bound. However, there is still huge gap between a practically effective compression method and its rigorous background of statistical learning theory. To resolve this issue, we develop a new theoretical framework for model compression and propose a new pruning method called {\it spectral pruning} based on this framework. We define the ``degrees of freedom'' to quantify the intrinsic dimensionality of a model by using the eigenvalue distribution of the covariance matrix across the internal nodes and show that the compression ability is essentially controlled by this quantity. Moreover, we present a sharp generalization error bound of the compressed model and characterize the bias--variance tradeoff induced by the compression procedure. We apply our method to several datasets to justify our theoretical analyses and show the superiority of the the proposed method.

stat.ML↗