Category Archives: Papers

New Preprint: Contraction Clustering (RASTER): A Very Fast Big Data Algorithm for Sequential and Parallel Density-Based Clustering in Linear Time, Constant Memory, and a Single Pass

A preprint of our paper “Contraction Clustering (RASTER): A Very Fast Big Data Algorithm for Sequential and Parallel Density-Based Clustering in Linear Time, Constant Memory, and a Single Pass” is now available on arXiv. The abstract is reproduced below:

Title: Contraction Clustering (RASTER): A Very Fast Big Data
Algorithm for Sequential and Parallel Density-Based Clustering
in Linear Time, Constant Memory, and a Single Pass

Authors: Gregor Ulm, Simon Smith, Adrian Nilsson,
Emil Gustavsson, Mats Jirstrand

Abstract: Clustering is an essential data mining tool for
analyzing and grouping similar objects. In big data
applications, however, many clustering algorithms are
infeasible due to their high memory requirements and/or
unfavorable runtime complexity. In contrast, Contraction
Clustering (RASTER) is a single-pass algorithm for identifying
density-based clusters with linear time complexity. Due to its
favorable runtime and the fact that its memory requirements
are constant, this algorithm is highly suitable for big data
applications where the amount of data to be processed is huge.
It consists of two steps: (1) a contraction step which
projects objects onto tiles and (2) an agglomeration step
which groups tiles into clusters. This algorithm is extremely
fast in both sequential and parallel execution. In single-
threaded execution on a contemporary workstation, an 
implementation in Rust processes a batch of 500 million points 
with 1 million clusters in less than 50 seconds. The speedup 
due to parallelization is significant, amounting to a factor 
of around 4 on an 8-core machine. 

New Preprint: Active-Code Replacement in the OODIDA Data Analytics Platform

A preprint of our paper “Active-Code Replacement in the OODIDA Data Analytics Platform” is now available on arXiv. It describes a key feature of OODIDA, which is a distributed system for data analytics for the automotive industry, targeting a fleet of reference vehicles. With active-code reloading, it is possible to replace code for custom computations without taking any component of the system down. The abstract is reproduced below:

Active-Code Replacement in the OODIDA Data Analytics Platform
Gregor Ulm, Emil Gustavsson, Mats Jirstrand

OODIDA (On-board/Off-board Distributed Data Analytics) is a
platform for distributing and executing concurrent data analysis
tasks. It targets a fleet of reference vehicles in the
automotive industry and has a particular focus on rapid
prototyping. Its underlying message-passing infrastructure has
been implemented in Erlang/OTP, but the external applications
for user interaction and carrying out data analysis tasks use
a language-independent JSON interface. These applications are
primarily implemented in Python. A data analyst interacting with
OODIDA uses a Python library. The bulk of the data analytics
tasks are performed by clients (on-board), while a central server
performs supplementary tasks (off-board). OODIDA can be
automatically packaged and deployed, which necessitates restarting
parts of the system, or all of it. This is potentially disruptive.
To address this issue, we added the ability to execute
user-defined Python modules on both the client and the server,
which can be replaced without restarting any part of the system.
Modules can even be swapped between iterations of an ongoing
assignment. This facilitates use cases such as iterative A/B
testing of machine learning algorithms or deploying experimental
algorithms on-the-fly. Active-code replacement is a key feature
of our system as well as an example of interoperability between
a functional and a non-functional programming language.

New Preprint: OODIDA: On-board/Off-board Distributed Data Analytics for Connected Vehicles

A preprint of our paper “OODIDA: On-board/Off-board Distributed Data Analytics for Connected Vehicles” is now available on arXiv. It describes a distributed system for data analytics for the automotive industry, targeting a fleet of reference vehicles. The abstract is reproduced below:

OODIDA: On-board/Off-board Distributed Data Analytics
for Connected Vehicles
Gregor Ulm, Emil Gustavsson, and Mats Jirstrand

Connected vehicles may produce gigabytes of data per
hour, which makes centralized data processing
impractical at the fleet level. In addition, there
are the problems of distributing tasks to edge
devices and processing them efficiently. Our solution
to this problem is OODIDA (On-board/off-board
Distributed Data Analytics), which is a platform that
tackles both task distribution to connected vehicles
as well as concurrent execution of large-scale tasks
on arbitrary subsets of clients. Its message-passing
infrastructure has been implemented in Erlang/OTP,
while the end points are language-agnostic. OODIDA is
highly scalable and able to process a significant
volume of data on resource-constrained clients.

Paper “Functional Federated Learning in Erlang (ffl-erl)” accepted for publication

Our paper “Functional Federated Learning in Erlang (fflerl)” has been accepted for publication in the Proceedings of the 26th International Workshop on Functional and (Constraint) Logic Programming (WFLP 2018). These proceedings will appear as Springer Lecture Notes in Computer Science Vol. 11285.

The abstract is below:

Functional Federated Learning in Erlang (ffl-erl) 
Gregor Ulm, Emil Gustavsson, and Mats Jirstrand

The functional programming language Erlang is well-suited
for concurrent and distributed applications, but numerical
computing is not seen as one of its strengths. Yet, the
recent introduction of Federated Learning, which leverages
client devices for decentralized machine learning tasks,
while a central server updates and distributes a global
model, motivated us to explore how well Erlang is suited to
that problem. We present the Federated Learning framework
ffl-erl and evaluate it in two scenarios: one in which the
entire system has been written in Erlang, and another in
which Erlang is relegated to coordinating client processes
that rely on performing numerical computations in the
programming language C. There is a concurrent as well as a
distributed implementation of each case. We show that Erlang
incurs a performance penalty, but for certain use cases this
may not be detrimental, considering the trade-off between
speed of development (Erlang) versus performance (C). Thus,
Erlang may be a viable alternative to C for some practical
machine learning tasks.

Accepted Paper at DIDL 2018

Our paper “A Performance Evaluation of Federated Learning Algorithms” has been accepted at the Second Workshop on Distributed Infrastructures for Deep Learning (DIDL 2018), which is colocated with the 2018 ACM/IFIP International Middleware Conference (Middleware 2018). This conference will take place from December 10 to 14 in Rennes, France. The abstract is reproduced below.

Title:
A Performance Evaluation of Federated Learning Algorithms
Adrian Nilsson, Simon Smith, Gregor Ulm, Emil Gustavsson, Mats Jirstrand (Fraunhofer-Chalmers Centre & Fraunhofer Center for Machine Learning)

Abstract:
Federated learning proposes an environment for distributed machine learning where a global model is learned by aggregating models that have been trained locally on data generating clients. Contrary to centralized optimization, clients can be very large in number and are characterized by challenges of data and network heterogeneity. Examples of clients include smartphones and connected vehicles, which highlights the practical relevance of this approach to distributed machine learning. We compare three algorithms for federated learning and benchmark their performance against a centralized approach where data resides on the server. The algorithms covered are Federated Averaging (FedAvg), Federated Stochastic Variance Reduced Gradient, and CO-OP. They are evaluated on the MNIST dataset using both i.i.d. and non-i.i.d. partitionings of the data. Our results show that, among the three federated algorithms, FedAvg achieves the highest accuracy, regardless of how data was partitioned. Our comparison between FedAvg and centralized learning shows that they are practically equivalent when i.i.d. data is used, but the centralized approach outperforms FedAvg with non-i.i.d. data.