Exploring a Bilingual Next Word Predictor for a Federated Learning Mobile Application

We explore a bilingual next-word predictor (NWP) under federated optimization for a mobile application. A character-based LSTM is server-trained on English and Dutch texts from a custom parallel corpora. This is used as the target performance. We simulate a federated learning environment to assess the feasibility of distributed training for the same model. The popular Federated Averaging (FedAvg) algorithm is used as the aggregation method. We show that the federated LSTM achieves decent performance, yet it is still sub-optimal. We suggest possible next steps to bridge this performance gap. Furthermore, we explore the effects of language imbalance varying the ratio of English and Dutch training texts (or clients). We show the model upholds performance (of the balanced case) up and until a 80/20 imbalance before decaying rapidly. Lastly, we describe the implementation of local client training, word prediction and client-server communication in a custom virtual keyboard for Android platforms. Additionally, homomorphic encryption is applied to provide with secure aggregation guarding the user from malicious servers.

Exploring a Bilingual Next Word Predictor for a Federated Learning Mobile Application

We explore a bilingual next-word predictor (NWP) under federated optimization for a mobile application. A character-based LSTM is server-trained on English and Dutch texts from a custom parallel corpora. This is used as the target performance. We simulate a federated learning environment to assess the feasibility of distributed training for the same model. The popular Federated Averaging (FedAvg) algorithm is used as the aggregation method. We show that the federated LSTM achieves decent performance, yet it is still sub-optimal. We suggest possible next steps to bridge this performance gap. Furthermore, we explore the effects of language imbalance varying the ratio of English and Dutch training texts (or clients). We show the model upholds performance (of the balanced case) up and until a 80/20 imbalance before decaying rapidly. Lastly, we describe the implementation of local client training, word prediction and client-server communication in a custom virtual keyboard for Android platforms. Additionally, homomorphic encryption is applied to provide with secure aggregation guarding the user from malicious servers.

Systematic comparison of ranking aggregation methods for gene lists in experimental results

A common experimental output in biomedical science is a list of genes implicated in a given biological process or disease. The results of a group of studies answering the same, or similar, questions can be combined by meta-analysis to find a consensus or a more reliable answer. Ranking aggregation methods can be used to combine gene lists from various sources in meta-analyses. Evaluating a ranking aggregation method on a specific type of dataset before using it is required to support the reliability of the result since the property of a dataset can influence the performance of an algorithm. Evaluation of aggregation methods is usually based on a simulated database especially for the algorithms designed for gene lists because of the lack of a known truth for real data. However, simulated datasets tend to be too small compared to experimental data and neglect key features, including heterogeneity of quality, relevance and the inclusion of unranked lists. In this study, a group of existing methods and their variations which are suitable for meta-analysis of gene lists are compared using simulated and real data. Simulated data was used to explore the performance of the aggregation methods as a function of emulating the common scenarios of real genomics data, with various heterogeneity of quality, noise level, and a mix of unranked and ranked data using 20000 possible entities. In addition to the evaluation with simulated data, a comparison using real genomic data on the SARS-CoV-2 virus, cancer (NSCLC), and bacteria (macrophage apoptosis) was performed. We summarise our evaluation results in terms of a simple flowchart to select a ranking aggregation method for genomics data.

Aggregation Method of Distributed Load Resources Based on Non-intrusive Load Identification

Because the power system contains a large number of user-side adjustable load resources, it can effectively enhance the operational flexibility of the power system and realize the safe, economical and efficient operation of the power grid by aggregating and modeling all kinds of resources and participating in the interactive response of the system as a whole. In this paper, a user-side adjustable load resource aggregation method based on non-intrusive load identification is proposed, which aims to obtain the load response potential of various users without intruding into the users, thus providing important support for power grid dispatching. Specifically, starting from the basic attributes of electrical equipment, considering the influence of numerical features such as current, harmonics, power, and V-I trajectory image features on load identification, the deep learning algorithm is used to deeply fuse the numerical features and image features in high-dimensional space, and then the fused advanced features are supervised by the Softmax classification algorithm, so as to effectively identify different types of electrical equipment. Finally, a bottom-up aggregation strategy is adopted to aggregate and model all kinds of load resources under the same station, so as to realize the accurate evaluation of the response ability of station resources. The simulation results of a numerical example verify the correctness and effectiveness of the proposed method.

Reasoning with PCP-Nets

We introduce PCP-nets, a formalism to model qualitative conditional preferences with probabilistic uncertainty. PCP-nets generalise CP-nets by allowing for uncertainty over the preference orderings. We define and study both optimality and dominance queries in PCP-nets, and we propose a tractable approximation of dominance which we show to be very accurate in our experimental setting. Since PCP-nets can be seen as a way to model a collection of weighted CP-nets, we also explore the use of PCP-nets in a multi-agent context, where individual agents submit CP-nets which are then aggregated into a single PCP-net. We consider various ways to perform such aggregation and we compare them via two notions of scores, based on well known voting theory concepts. Experimental results allow us to identify the aggregation method that better represents the given set of CP-nets and the most efficient dominance procedure to be used in the multi-agent context.

Federated Learning with Random Communication and Dynamic Aggregation

Federated Learning (FL) is a setting that allows clients to train a joint global model collaboratively while keeping data locally. Due to FL has advantages of data confidential and distributed computing, interest in this area has increased. In this paper, we designed a new FL algorithm named FedRAD. Random communication and dynamic aggregation methods are proposed for FedRAD. Random communication method enables FL system use the combination of fixed communication interval and constrained variable intervals in a single task. Dynamic aggregation method reforms aggregation weights and makes weights update automately. Both methods aim to improve model performance. We evaluated two proposed methods respectively, and compared FedRAD with three algorithms on three hyperparameters. Results at CIFAR-10 demonstrate that each method has good performance, and FedRAD can achieve higher classification accuracy than state-of-the-art FL algorithms.

A Survey of Dementia Knowledge and Recognition of Dementia Prevention and Practice in Healthy Older Adults

<b><i>Aim:</i></b> The aim of this study was to investigate the level of dementia knowledge and behaviors recognized as dementia-preventive and the practice thereof among healthy older adults who are highly motivated to engage in activities. <b><i>Methods:</i></b> The participants were older adults registered at the Silver Human Resource Center of city A, and participants anonymously filled questionnaires through the aggregation method in January 2020. <b><i>Results:</i></b> The analysis included 78 participants (the effective response rate was 49.7%). All participants were aware of at least 4 dementia-preventive behaviors, and about 80% of all participants practiced at least 1 preventive behavior. Approximately 20% of participants were not practicing preventive behaviors at all. The elderly aged 65 to 74 years had more knowledge about dementia and more types of behavior perceived as dementia-preventive than the elderly aged 75 years and older. There was no significant correlation between the level of dementia knowledge and the number of types of dementia-preventive behaviors. <b><i>Conclusions:</i></b> Healthy older adults were aware of numerous behaviors for dementia prevention, and most older adults practiced preventive behaviors. In contrast, even with a high amount of knowledge about dementia, a small number of healthy older adults did not translate this knowledge into preventative behavioral practices. High levels of dementia knowledge do not translate into preventive behavioral practices.