In machine learning applications over Big streaming Data, Neural Networks (NNs) are continuously and rapidly trained over voluminous data arriving at high speeds. As soon as a new version of the NN becomes available, it gets deployed for prediction purposes (e.g. classification). The real-time character of such applications greatly depends on the volume and velocity of the data streams, as well as the NN complexity. Training on large volume of ingested streams or using complex NNs, potentially increases accuracy, but may compromise the real-time character of those applications. In this work, we present SuBiTO, a framework that automatically and continuously learns the training time vs accuracy trade-offs as new data stream in and fine tunes: (i) the number, size and type of NN layers; (ii) the size of the ingested data via stream synopses specific parameters; and (iii) the number of training epochs. Finally, SuBiTO suggests optimal sets of such parameters and detects concept drifts, enabling the human operator adapt these parameters on-the-fly, at runtime.

Many IoT applications from diverse domains rely on real-time, online analytics workflow execution to timely support decision making procedures. The efficient execution of analytics workflows requires the utilization of the processing power available across the cloud to edge continuum. Nonetheless, suggesting the optimal workflow execution over a large network of heterogeneous devices is a challenging task. The increased IoT network size increases the complexity of the optimization problem at hand. The ingested data streams exhibit highly volatile properties. The population of network devices dynamically changes. We introduce DAG*, an A*-alike algorithm that prunes large amounts of the search space explored for suggesting the most efficient workflow execution with formal optimality guarantees. We provide an incremental version of DAG* retaining the optimality property. Our experimentation in real-world scenarios shows that DAG* suggests the optimal workflow execution with 3 to 31 orders of magnitude fewer iterations compared to the entire search space size, outperforming heuristics employed in prior state of the art up to x4.S wrt the goodness of the suggested workflow.

In this work, we introduce RATS ( R esource A llocator for T umor S imulations), the first optimizer for the execution of tumor simulations over HPC infrastructures. Given a set of drug therapies under in-silico study, the optimization framework of RATS can: (i) devise the optimal number of cores and prescribe the required number of core hours; and (ii) under core capacity constraints, RATS schedules the execution of simulations minimizing the overall number of core hours, simultaneously prioritizing the execution of expectedly promising in-silico trials higher compared to unpromising ones. RATS is deployed by life scientists at the Barcelona Supercomputing Center to remove the burden of blindly guessing the core hours needing to be reserved from HPC admins to study various tumor treatment methodologies, as well as to rapidly distinguish effective drug combinations, thus, potentially cutting time to market for new cancer therapies. The latter is further elevated by the RATS+ extension we plug into the initial framework. RATS+ employs a Transfer Learning approach to leverage optimization models and decisions from prior in-silico studies, thereby reducing the optimization effort required for new studies in this domain. Our experimental evaluation, on real-world data derived from the execution of more than 2500 tumor simulations on the MareNostrum4 supercomputer, confirms the effectiveness of both RATS and RATS+ across the aforementioned performance dimensions.

In recent years, popular touristic destinations face overtourism. Local communities suffer from its consequences in several ways. Among others, overpricing and profiteering harms local societies and economies deeply. In this paper we focus on the problem of determining fair hotel room prices. Specifically, we put forward a dynamic pricing policy where the price of a room depends not only on the demand of the hotel it belongs to but also on the demand of: (i) similar rooms in the area and (ii) their hotels. To this purpose, we model our setting as a cooperative game and exploit an appropriate game theoretic solution concept that promotes fairness both on the customers' and the providers' side. Our simulation results involving price adjustments across real-world hotels datasets, confirm that ours is a fair dynamic pricing policy, avoiding both over- and under-pricing hotel rooms.

We introduce RATS (Resource Allocator for Tumor Simulations), the first optimizer for the execution of tumor simulations over HPC infrastructures. The optimization framework of RATS incorporates 3 vital performance criteria (i) expected utility of a simulation in terms of effective drug combination on the simulated tumor, (ii) simulation execution time and (iii) number of cores required for achieving that execution time. RATS is to be used by life scientists at the Barcelona Supercomputing Center to not only remove the burden of blindly guessing the core hours we need to reserve from HPC admins to study various tumor treatment methodologies, but also to help in more rapidly distinguishing effective drug combinations, thus, potentially cutting time to market for new cancer therapies.