Search

Federated learning (FL) has emerged as a transformative training paradigm, particularly invaluable in privacy-sensitive domains like healthcare. However, client heterogeneity in data, computing power, and tasks poses a significant challenge. To address such a challenge, we propose an FL optimization algorithm that integrates model delta regularization, personalized models, federated knowledge distillation, and mix-pooling. Model delta regularization optimizes model updates centrally on the server, efficiently updating clients with minimal communication costs. Personalized models and federated knowledge distillation strategies are employed to tackle task heterogeneity effectively. Additionally, mix-pooling is introduced to accommodate variations in the sensitivity of readout operations. Experimental results demonstrate the remarkable accuracy and rapid convergence achieved by model delta regularization. Additionally, the federated knowledge distillation algorithm notably improves FL performance, especially in scenarios with diverse data. Moreover, mix-pooling readout operations provide tangible benefits for clients, showing the effectiveness of our proposed methods.

Deep generative models, such as generative adversarial networks (GANs), have been employed for $de~novo$ molecular generation in drug discovery. Most prior studies have utilized reinforcement learning (RL) algorithms, particularly Monte Carlo tree search (MCTS), to handle the discrete nature of molecular representations in GANs. However, due to the inherent instability in training GANs and RL models, along with the high computational cost associated with MCTS sampling, MCTS RL-based GANs struggle to scale to large chemical databases. To tackle these challenges, this study introduces a novel GAN based on actor-critic RL with instant and global rewards, called InstGAN, to generate molecules at the token-level with multi-property optimization. Furthermore, maximized information entropy is leveraged to alleviate the mode collapse. The experimental results demonstrate that InstGAN outperforms other baselines, achieves comparable performance to state-of-the-art models, and efficiently generates molecules with multi-property optimization. The source code will be released upon acceptance of the paper.

We consider a collection of $k \geq 2$ robots that evolve in a ring-shaped network without common orientation, and address a variant of the crash-tolerant gathering problem called the \emph{Stand-Up Indulgent Gathering} (SUIG): given a collection of robots, if no robot crashes, robots have to meet at the same arbitrary location, not known beforehand, in finite time; if one robot or more robots crash on the same location, the remaining correct robots gather at the location of the crashed robots. We aim at characterizing the solvability of the SUIG problem without multiplicity detection capability.

We consider a strong variant of the crash fault-tolerant gathering problem called stand-up indulgent gathering (SUIG), by robots endowed with limited visibility sensors and lights on line-shaped networks. In this problem, a group of mobile robots must eventually gather at a single location, not known beforehand, regardless of the occurrence of crashes. Differently from previous work that considered unlimited visibility, we assume that robots can observe nodes only within a certain fixed distance (that is, they are myopic), and emit a visible color from a fixed set (that is, they are luminous), without multiplicity detection. We consider algorithms depending on two parameters related to the initial configuration: $M_{init}$, which denotes the number of nodes between two border nodes, and $O_{init}$, which denotes the number of nodes hosting robots. Then, a border node is a node hosting one or more robots that cannot see other robots on at least one side. Our main contribution is to prove that, if $M_{init}$ or $O_{init}$ is odd, SUIG can be solved in the fully synchronous model.

The successful application of large pre-trained models such as BERT in natural language processing has attracted more attention from researchers. Since the BERT typically acts as an end-to-end black box, classification systems based on it usually have difficulty in interpretation and low robustness. This paper proposes a visual interpretation-based self-improving classification model with a combination of virtual adversarial training (VAT) and BERT models to address the above problems. Specifically, a fine-tuned BERT model is used as a classifier to classify the sentiment of the text. Then, the predicted sentiment classification labels are used as part of the input of another BERT for spam classification via a semi-supervised training manner using VAT. Additionally, visualization techniques, including visualizing the importance of words and normalizing the attention head matrix, are employed to analyze the relevance of each component to classification accuracy. Moreover, brand-new features will be found in the visual analysis, and classification performance will be improved. Experimental results on Twitter's tweet dataset demonstrate the effectiveness of the proposed model on the classification task. Furthermore, the ablation study results illustrate the effect of different components of the proposed model on the classification results.

We consider a variant of the crash-fault gathering problem called stand-up indulgent gathering (SUIG). In this problem, a group of mobile robots must eventually gather at a single location, which is not known in advance. If no robots crash, they must all meet at the same location. However, if one or more robots crash at a single location, all non-crashed robots must eventually gather at that location. The SUIG problem was first introduced for robots operating in a two-dimensional continuous Euclidean space, with most solutions relying on the ability of robots to move a prescribed (real) distance at each time instant. In this paper, we investigate the SUIG problem for robots operating in a discrete universe (i.e., a graph) where they can only move one unit of distance (i.e., to an adjacent node) at each time instant. Specifically, we focus on line-shaped networks and characterize the solvability of the SUIG problem for oblivious robots without multiplicity detection.

We consider the gathering problem for asynchronous and oblivious robots that cannot communicate explicitly with each other, but are endowed with visibility sensors that allow them to see the positions of the other robots. Most of the investigations on the gathering problem on the discrete universe are done on ring shaped networks due to the number of symmetric configuration. We extend in this paper the study of the gathering problem on torus shaped networks assuming robots endowed with local weak multiplicity detection. That is, robots cannot make the difference between nodes occupied by only one robot from those occupied by more than one robots unless it is their current node. As a consequence, solutions based on creating a single multiplicity node as a landmark for the gathering cannot be used. We present in this paper a deterministic algorithm that solves the gathering problem starting from any rigid configuration on an asymmetric unoriented torus shaped network., Comment: 41 pages

We consider the problem of constructing a maximum independent set with mobile myopic luminous robots on a grid network whose size is finite but unknown to the robots. In this setting, the robots enter the grid network one-by-one from a corner of the grid, and they eventually have to be disseminated on the grid nodes so that the occupied positions form a maximum independent set of the network. We assume that robots are asynchronous, anonymous, silent, and they execute the same distributed algorithm. In this paper, we propose two algorithms: The first one assumes the number of light colors of each robot is three and the visible range is two, but uses additional strong assumptions of port-numbering for each node. To delete this assumption, the second one assumes the number of light colors of each robot is seven and the visible range is three. In both algorithms, the number of movements is $O(n(L+l))$ steps where $n$ is the number of nodes and $L$ and $l$ are the grid dimensions.

In this paper, we present two self-stabilizing algorithms that enable a single (mobile) agent to explore graphs. The agent visits all nodes starting from any configuration, \ie regardless of the initial state of the agent, the initial states of all nodes, and the initial location of the agent. We evaluate the algorithms using two metrics: cover time, which is the number of moves required to visit all nodes, and memory usage, which includes the storage needed for the state of the agent and the state of each node. The first algorithm is randomized. Given an integer $c = \Omega(n)$, the cover time of this algorithm is optimal, \ie $O(m)$ in expectation, and the memory requirements for the agent and each node $v$ are $O(\log c)$ and $O(\log (c+\delta_v))$ bits, respectively, where $n$ and $m$ are the numbers of nodes and edges, respectively, and $\delta_v$ is the degree of $v$. The second algorithm is deterministic. It requires an input integer $k \ge \max(D,d_{\mathrm{max}})$, where $D$ and $d_{\mathrm{max}}$ are the diameter and the maximum degree of the graph, respectively. The cover time of this algorithm is $O(m + nD)$, and it uses $O(\log k)$ bits both for agent memory and each node.

A metamorphic robotic system (MRS) is composed of anonymous, memoryless, and autonomous modules that execute an identical distributed algorithm to move while keeping the connectivity of the modules. For an MRS, the number of modules required to solve a given task is an important complexity measure. Here, we consider evacuation from a finite two-dimensional square grid field by an MRS. This study aims to establish the minimum number of modules required to solve the evacuation problem under several conditions. We consider a rectangular field surrounded by walls with at least one exit. Our results show that two modules are necessary and sufficient for evacuation from any rectangular field if equipped with a global compass, which provides the modules with a common sense of direction. After that, we focus on the case of modules without a global compass and show that four (resp. seven) modules are necessary and sufficient for restricted (resp. any) initial shapes of an MRS. We also show that two modules are sufficient when an MRS is touching a wall in an initial configuration. Then, we clarify the condition to stop an MRS after evacuation of a rectangular field. Finally, we extend these results to mazes and convex fields.

We investigate gathering algorithms for asynchronous autonomous mobile robots moving in uniform ring-shaped networks. Different from most work using the Look-Compute-Move (LCM) model, we assume that robots have limited visibility and lights. That is, robots can observe nodes only within a certain fixed distance, and emit a color from a set of constant number of colors. We consider gathering algorithms depending on two parameters related to the initial configuration: $M_{init}$, which denotes the number of nodes between two border nodes, and $O_{init}$, which denotes the number of nodes hosting robots between two border nodes. In both cases, a border node is a node hosting one or more robots that cannot see other robots on at least one side. Our main contribution is to prove that, if $M_{init}$ or $O_{init}$ is odd, gathering is always feasible with three or four colors. The proposed algorithms do not require additional assumptions, such as knowledge of the number of robots, multiplicity detection capabilities, or the assumption of towerless initial configurations. These results demonstrate the power of lights to achieve gathering of robots with limited visibility., Comment: This is a full version of the conference paper in OPODIS2019

Local mutual exclusion guarantees that no two neighboring processes enter a critical section at the same time while satisfying both mutual exclusion and no starvation properties. On the other hand, processes may want to execute some operation simultaneously with the neighbors. Of course, we can use a globally synchronized clock to achieve the task but it is very expensive to realize it in a distributed system in general. In this paper, we define a new concept neighborhood mutual remainder. A distributed algorithm that satisfies the neighborhood mutual remainder requirement should satisfy global fairness, l-exclusion and repeated local rendezvous requirements. Global fairness is satisfied when each process (that requests to enter the critical section infinitely often) executes the critical section infinitely often, l-exclusion is satisfied when at most l neighboring processes enter the critical section at the same time, and repeated local rendezvous is satisfied when for each process infinitely often no process in the closed neighborhood is in the critical or trying section. We first formalize the concept of neighborhood mutual remainder, and give a simple self-stabilizing algorithm to demonstrate the design paradigm to achieve neighborhood mutual remainder. We also present two applications of neighborhood mutual remainder to a Look-Compute-Move robot system. One is for implementing a move-atomic property and the other is for implementing FSYNC scheduler, where robots possess an independent clock that is advanced in the same speed. These are the first self-stabilizing implementations of the LCM synchronization., Comment: 16 pages

This paper considers the global $(l,k)$-CS problem which is the problem of controlling the system in such a way that, at least $l$ and at most $k$ processes must be in the CS at a time in the network. In this paper, a distributed solution is proposed in the asynchronous message-passing model. Our solution is a versatile composition method of algorithms for $l$-mutual inclusion and $k$-mutual exclusion. Its message complexity is $O(|Q|)$, where $|Q|$ is the maximum size for the quorum of a coterie used by the algorithm, which is typically $|Q| = \sqrt{n}$., Comment: This is a modified version of the conference paper in PDAA2017

We propose a gathering protocol for an even number of robots in a ring-shaped network that allows symmetric but not periodic configurations as initial configurations, yet uses only local weak multiplicity detection. Robots are assumed to be anonymous and oblivious, and the execution model is the non- atomic CORDA model with asynchronous fair scheduling. In our scheme, the number of robots k must be greater than 8, the number of nodes n on a network must be odd and greater than k+3. The running time of our protocol is O(n2) asynchronous rounds., Comment: arXiv admin note: text overlap with arXiv:1104.5660

We consider a set of k autonomous robots that are endowed with visibility sensors (but that are otherwise unable to communicate) and motion actuators. Those robots must collaborate to reach a sin- gle vertex that is unknown beforehand, and to remain there hereafter. Previous works on gathering in ring-shaped networks suggest that there exists a tradeoff between the size of the set of potential initial configurations, and the power of the sensing capabilities of the robots (i.e. the larger the initial configuration set, the most powerful the sensor needs to be). We prove that there is no such trade off. We propose a gathering protocol for an odd number of robots in a ring-shaped network that allows symmetric but not periodic configurations as initial configurations, yet uses only local weak multiplicity detection. Robots are assumed to be anonymous and oblivious, and the execution model is the non-atomic CORDA model with asynchronous fair scheduling. Our protocol allows the largest set of initial configurations (with respect to impossibility results) yet uses the weakest multiplicity detector to date. The time complexity of our protocol is O(n2), where n denotes the size of the ring. Compared to previous work that also uses local weak multiplicity detection, we do not have the constraint that k < n/2 (here, we simply have 2 < k < n - 3).