Search

The use of machine learning for time series prediction has become increasingly popular across various industries thanks to the availability of time series data and advancements in machine learning algorithms. However, traditional methods for time series forecasting rely on pre-optimized models that are ill-equipped to handle unpredictable patterns in data. In this paper, we present EAMDrift, a novel method that combines forecasts from multiple individual predictors by weighting each prediction according to a performance metric. EAMDrift is designed to automatically adapt to out-of-distribution patterns in data and identify the most appropriate models to use at each moment through interpretable mechanisms, which include an automatic retraining process. Specifically, we encode different concepts with different models, each functioning as an observer of specific behaviors. The activation of the overall model then identifies which subset of the concept observers is identifying concepts in the data. This activation is interpretable and based on learned rules, allowing to study of input variables relations. Our study on real-world datasets shows that EAMDrift outperforms individual baseline models by 20% and achieves comparable accuracy results to non-interpretable ensemble models. These findings demonstrate the efficacy of EAMDrift for time-series prediction and highlight the importance of interpretability in machine learning models., Comment: Submitted to ECML PKDD 2023

In this technical report we compare different deep learning models for prediction of water depth rasters at high spatial resolution. Efficient, accurate, and fast methods for water depth prediction are nowadays important as urban floods are increasing due to higher rainfall intensity caused by climate change, expansion of cities and changes in land use. While hydrodynamic models models can provide reliable forecasts by simulating water depth at every location of a catchment, they also have a high computational burden which jeopardizes their application to real-time prediction in large urban areas at high spatial resolution. Here, we propose to address this issue by using data-driven techniques. Specifically, we evaluate deep learning models which are trained to reproduce the data simulated by the CADDIES cellular-automata flood model, providing flood forecasts that can occur at different future time horizons. The advantage of using such models is that they can learn the underlying physical phenomena a priori, preventing manual parameter setting and computational burden. We perform experiments on a dataset consisting of two catchments areas within Switzerland with 18 simpler, short rainfall patterns and 4 long, more complex ones. Our results show that the deep learning models present in general lower errors compared to the other methods, especially for water depths $>0.5m$. However, when testing on more complex rainfall events or unseen catchment areas, the deep models do not show benefits over the simpler ones.

Web3 aims at creating a decentralized platform that is competitive with modern cloud infrastructures that support today's Internet. However, Web3 is still limited, supporting only applications in the domains of content creation and sharing, decentralized financing, and decentralized communication. This is mainly due to the technologies supporting Web3: blockchain, IPFS, and libp2p, that although provide a good collection of tools to develop Web3 applications, are still limited in terms of design and performance. This motivates the need to better understand these technologies as to enable novel optimizations that can push Web3 to its full potential. Unfortunately, understanding the current behavior of a fully decentralized large-scale distributed system is a difficult task, as there is no centralized authority that has full knowledge of the system operation. To this end, in this paper we characterize the workload of IPFS, a key enabler of Web3. To achieve this, we have collected traces from accesses performed by users to one of the most popular IPFS gateways located in North America for a period of two weeks. Through the fine analysis of these traces, we gathered the amount of requests to the system, and found the providers of the requested content. With this data, we characterize both the popularity of requested and provided content, as well as their geo-location (by matching IP address with the MaxMind database). Our results show that most of the requests in IPFS are only to a few different content, that is provided by large portion of peers in the system. Furthermore, our analysis also shows that most requests are provided by the two largest portions of providers in the system, located in North America and Europe. With these insights, we conclude that the current IPFS architecture is sub-optimal and propose a research agenda for the future.

Prototyping and implementing distributed algorithms, particularly those that address challenges related with fault-tolerance and dependability, is a time consuming task. This is, in part, due to the need of addressing low level aspects such as management of communication channels, controlling timeouts or periodic tasks, and dealing with concurrency issues. This has a significant impact for researchers that want to build prototypes for conducting experimental evaluation; practitioners that want to compare different design alternatives/solutions; and even for practical teaching activities on distributed algorithms courses. In this paper we present Babel, a novel framework to develop, implement, and execute distributed protocols and systems. Babel promotes an event driven programming and execution model that simplifies the task of translating typical specifications or descriptions of algorithms into performant prototypes, while allowing the programmer to focus on the relevant challenges of these algorithms by transparently handling time consuming low level aspects. Furthermore, Babel provides, and allows the definition of, networking components that can capture different network capabilities (e.g., P2P, Client/Server, phi-accrual Failure Detector), making the code mostly independent from the underlying communication aspects. Babel was built to be generic and can be used to implement a wide variety of different classes of distributed protocols. We conduct our experimental work with two relevant case studies, a Peer-to-Peer application and a State Machine Replication application, that show the generality and ease of use of Babel and present competitive performance when compared with significantly more complex implementations., Comment: 18 pages, 2 figures

An increasing number of mobile applications share location-dependent information, from collaborative applications and social networks to location-based games. For such applications, peer-to-peer architectures where mobile devices share information directly may grant lower latency and reduce server load. In this work, we propose a framework to support these applications, providing location-dependent replication. Data has an associated location and is replicated in the mobile devices that show interest and are close to that location. Our proposal combines peer-to-peer synchronisation among mobile devices and replication at edge servers. Our preliminary results are promising., Comment: 9 pages, 4 figures and 3 listings

The concurrency control algorithms in transactional systems limits concurrency to provide strong semantics, which leads to poor performance under high contention. As a consequence, many transactional systems eschew strong semantics to achieve acceptable performance. We show that by leveraging semantic information associated with the transactional programs to increase concurrency, it is possible to significantly improve performance while maintaining linearizability. To this end, we introduce the lazy state determination API to easily expose the semantics of application transactions to the database, and propose new optimistic and pessimistic concurrency control algorithms that leverage this information to safely increase concurrency in the presence of contention. Our evaluation shows that our approach can achieve up to 5x more throughput with 1.5c less latency than standard techniques in the popular TPC-C benchmark.

Computational complexity has been the bottleneck of applying physically-based simulations on large urban areas with high spatial resolution for efficient and systematic flooding analyses and risk assessments. To address this issue of long computational time, this paper proposes that the prediction of maximum water depth rasters can be considered as an image-to-image translation problem where the results are generated from input elevation rasters using the information learned from data rather than by conducting simulations, which can significantly accelerate the prediction process. The proposed approach was implemented by a deep convolutional neural network trained on flood simulation data of 18 designed hyetographs on three selected catchments. Multiple tests with both designed and real rainfall events were performed and the results show that the flood predictions by neural network uses only 0.5 % of time comparing with physically-based approaches, with promising accuracy and ability of generalizations. The proposed neural network can also potentially be applied to different but relevant problems including flood predictions for urban layout planning.