Training deep learning (DL) models often takes a significant amount of time and is thus typically performed on expensive GPUs to speed up the process. However, data loading has recently been identified as one of the main performance bottlenecks in DL, resulting in GPU under-utilization. Looking forward, the combination of larger datasets and faster GPUs will exacerbate the problem. The data management community has started to address this by proposing data loading optimization techniques, including lossy image compression. While lossy compression is a conceptually promising approach for mitigating data loading bottlenecks in DL, there is only limited understanding of its efficacy in terms of impact on model throughput and accuracy. In this paper, we present an extensive experimental analysis of lossy image compression as a means to improve the performance of neural network training. We find that lossy compression can improve both throughput and accuracy of DL pipelines if resources such as time or storage capacity are limited. Furthermore, the choice of compression quality and codec are important hyperparameters when training deep neural networks.

Many companies and businesses train and deploy machine learning (ML) pipelines to answer prediction queries. In many applications, new training data continuously becomes available. A typical approach to ensure that ML models are up-to-date is to retrain the ML pipelines following a schedule, e.g., every day on the last seven days of data. Several use cases, such as A/B testing and ensemble learning, require many pipelines to be deployed in parallel. Existing solutions train each pipeline separately, which generates redundant data processing. Our goal is to eliminate redundant data processing in such scenarios using materialization and reuse optimizations. Our solution comprises of two main parts. First, we propose a materialization algorithm that given a storage budget, materializes the subset of the artifacts to minimize the run time of the subsequent executions. Second, we design a reuse algorithm to generate an execution plan by combining the pipelines into a directed acyclic graph (DAG) and reusing the materialized artifacts when appropriate. Our experiments show that our solution can reduce the training time by up to an order of magnitude for different deployment scenarios.

Data science workflows are largely exploratory, dealing with under-specified objectives, open-ended problems, and unknown business value. Therefore, little investment is made in systematic acquisition, integration, and pre-processing of data. This lack of infrastructure results in redundant manual effort and computation. Furthermore, central data consolidation is not always technically or economically desirable or even feasible (e.g., due to privacy, and/or data ownership). The ExDRa system aims to provide system infrastructure for this exploratory data science process on federated and heterogeneous, raw data sources. Technical focus areas include (1) ad-hoc and federated data integration on raw data, (2) data organization and reuse of intermediates, and (3) optimization of the data science lifecycle, under awareness of partially accessible data. In this paper, we describe use cases, the overall system architecture, selected features of SystemDS' new federated backend (for federated linear algebra programs, federated parameter servers, and federated data preparation), as well as promising initial results. Beyond existing work on federated learning, ExDRa focuses on enterprise federated ML and related data pre-processing challenges. In this context, federated ML has the potential to create a more fine-grained spectrum of data ownership and thus, even new markets.

Effective collaboration among data scientists results in high-quality and efficient machine learning (ML) workloads. In a collaborative environment, such as Kaggle or Google Colabratory, users typically re-execute or modify published scripts to recreate or improve the result. This introduces many redundant data processing and model training operations. Reusing the data generated by the redundant operations leads to the more efficient execution of future workloads. However, existing collaborative environments lack a data management component for storing and reusing the result of previously executed operations. In this paper, we present a system to optimize the execution of ML workloads in collaborative environments by reusing previously performed operations and their results. We utilize a so-called Experiment Graph (EG) to store the artifacts, i.e., raw and intermediate data or ML models, as vertices and operations of ML workloads as edges. In theory, the size of EG can become unnecessarily large, while the storage budget might be limited. At the same time, for some artifacts, the overall storage and retrieval cost might outweigh the recomputation cost. To address this issue, we propose two algorithms for materializing artifacts based on their likelihood of future reuse. Given the materialized artifacts inside EG, we devise a linear-time reuse algorithm to find the optimal execution plan for incoming ML workloads. Our reuse algorithm only incurs a negligible overhead and scales for the high number of incoming ML workloads in collaborative environments. Our experiments show that we improve the run-time by one order of magnitude for repeated execution of the workloads and 50% for the execution of modified workloads in collaborative environments.

Several scientific domains operate on mesh data structures. Computations on meshes share a set of basic abstractions which relates to topological and geometric structure of meshes. Due to the complexity of modeling and querying meshes, the database community has neglected the need for a dedicated mesh database and in particular a high-level mesh query language. Currently, such computations is done by using general-purpose programming languages. Such implementations are tightly coupled with dataset characteristics and thus are less reusable, their maintenance cost is high, and their performance is limited to the memory size. In this paper, we describe syntax and semantic of a high-level query language for scientific meshes (called MQuery) with a level of abstraction that can express queries based on topological and geometric navigations. The language is build on top of a graph-based conceptual mesh data model. We report our results comparing two implementations of the language on top of relational and graph databases with a mesh library.

Although, many applications use unstructured meshes, there is no specialized mesh database which supports storing and querying mesh data. Existing mesh libraries do not support declarative querying and are expensive to maintain. A mesh database can benefit the domains in several ways such as: declarative query language, ease of maintenance, etc. In this paper, we propose the Incidence multi-Graph Complex (ImG-Complex) data model for storing topological aspects of meshes in a database. ImG-Complex extends incidence graph (IG) model with multi-incidence information to represent a new object class which we call ImG-Complexes. We introduce optional and application-specific constraints to limit the ImG model to smaller object classes and validate mesh structures based on the modeled object class properties. We show how Neo4j graph database can be used to query mesh topology based on the (possibly constrained) ImG model. Finally, we experiment Neo4j and PostgreSQL performance on executing topological mesh queries.

Despite ubiquitous usage of unstructured mesh in many application domains (e.g., computer aided design, scientific simulation, climate modeling, etc.), there is no specialized mesh database which supports storing and querying such data structures. Existing mesh libraries use file-based APIs which do not support declarative querying and are difficult to maintain. A mesh database can benefit these domains in several ways such as: declarative query language, ease of maintenance, query optimization, etc. In this thesis work, the core idea is to have a very general model which can represent objects from different domains and specialize it to smaller object classes using combinatorial constraints. We propose the Incidence multi-Graph Complex (ImG-Complex) data model for storing combinatorial aspect of meshes in a database. We extend incidence graph (IG) representation with multi-incidence information (ImG) to represent a class of objects which we call ImG-Complexes. ImG-Complex can support a wide range of application domains. We introduce optional and application-specific constraints to restrain the general ImG model to specific object classes or specific geometric representations. The constraints check validity of meshes based on the properties of the modeled object class. Finally, we show how graph databases can be utilized and reused to query some combinatorial mesh queries based on the (possibly constrained) ImG model. In particular, we show the strengths and limitations of a graph-only query language in expressing combinatorial mesh queries.

To understand and improve processes in organizations, six key questions need to be answered, namely, what, how, where, who, when, why. Organizations with established processes have IT system(s) that gather(s) information about some or all of the key questions. Software organizations usually have defined processes, but they usually lack information about how processes are actually executed. Moreover, there is no explicit information about process instances and activities. Existing process mining techniques face problems in coping with such environment. We propose a tool, Egidio, which uses non-invasively collected data and builds organizational models. In particular, we explain the tool within a software company, which is able to extract different aspects of development processes. The main contribution of Egidio is the ability to mine processes and organizational models from fine-grained data collected in a non-invasive manner, without interrupting the developers' work.