Last month, I successfully defended my Master’s thesis titled “MultiETSC: Automated Machine Learning for Time Series Classification”. I have been working on my MultiETSC project for a little over a year as a part of the ADA group, supervised by Can Wang, Mitra Baratchi and Holger H. Hoos. In this project, we developed an AutoML approach to the problem of early time series classification (ETSC).
ETSC is a problem that is relevant for time-critical applications of time series classifications. It arises from the desire to get classification output before the full time series is observed. ETSC requires a classifier to decide at what point to classify based on the partially observed time series. This decision trades off the earliness of the classification for the accuracy of the classification. Therefore, any ETSC algorithm is faced with two competing objectives: earliness and accuracy.
For MutliETSC, we took eight existing and one naive self-produced ETSC algorithms and partly rewrote them to share a common interface. On this set of algorithms, we applied the MO-ParamILS configurator, which can optimise multiple objectives simultaneously. MultiETSC produces a set of configurations that give multiple options for trading off earliness against accuracy, while non of these configurations is strictly better than the other. This allows a user to pick a configuration that fits the ETSC problem at hand.
In our research, we show that leaving out either the algorithm selection or the multi-objective optimisation parts will both result in inferior results. This means that both these aspects significantly contribute to the quality of MultiETSC’s output.
If you want to know more about this project, have a look at the project page.
Finishing this project, sadly means I will be leaving the ADA group, but I am happy to have had the opportunity to work in such a great environment with such smart and motivated people as in the ADA group.
It would seem that former Master student Laurens Arp will not be leaving the ADA group just yet, as he has just renewed his membership for the group, this time as a PhD student.
Prior to starting his PhD studies, Laurens worked on his Master’s Thesis on spatial interpolation within the ADA group supervised by dr. Mitra Baratchi and prof.dr. Holger Hoos, culminating in his MSc degree obtained in 2020.
The research Laurens will be carrying out now is part of dr. Mitra Baratchi’s “Physics-aware Spatio-temporal Machine Learning for Earth Observation Data” project, which involves a collaboration with the European Space Agency. In this project, the goal is to estimate environmental parameters (leaf density, CO2 levels in the atmosphere, etc.) from optical Earth observation data acquired by satellites. Generally speaking, there are two approaches one might take to this end. The first of those, traditionally favoured by the ecology community, would be to create inverted versions of deterministic radiative transfer models, which model how radiation (captured in Earth observation data) is affected by environmental parameters (ground truth). The other approach, perhaps more familiar to the AI and machine learning communities, would be to use spatio-temporal machine learning methods (e.g., temporal CNNs) to model the correlation between data (spectral images of radiation) and ground truth (environmental parameters). Both approaches have their own strengths and weaknesses, which motivates the development of hybrid models benefiting from the strengths of both. Thus, the creation of such hybrid models is what the project will aim to achieve.
Between global warming, the loss of biodiversity through mass extinction and other ecological and environmental perils the Earth is faced with these days, the challenges we, as residents of the Earth, are faced with can be daunting indeed. It is our hope that the models we will create will provide a meaningful contribution to the global efforts to address these issues.
Besides working in the ADA research group, Richard studies Computer Science & Economics where his main interest lies in being able to apply codes to real-life applications. Therefore, he found the ADA group an excellent environment to work on his bachelor project.
Richards’ project is concerned with Parallel Algorithm Portfolios, and adding this new functionality to the Sparkle platform. The problems which can be solved by the functionality will encompass decision problems, optimisation problems with a single solution and optimisation problems with several possible solution. All being well, the project will showcase a practical application of Parallel Portfolios.
This blog post has originally been published on automl.org
By Koen van der Blom, Holger Hoos, Alex Serban, Joost Visser
In our global survey among teams that build ML applications, we found ample room for increased adoption of AutoML techniques. While AutoML is adopted at least partially by more than 70% of teams in research labs and tech companies, for teams in non-tech and government organisations adoption is still below 50%. Full adoption remains limited to under 10% in research and tech, and to less than 5% in non-tech and government.
Software engineering for machine learning
With the inclusion of ML techniques in software, the risks and requirements of the software also changes. In turn, the best ways to maintain and develop software with ML components is different from traditional software engineering (TSE). We call this new field software engineering for machine learning (SE4ML).
Based on a study of the academic and “grey” literature (the latter comprises blog articles, presentation slides and white papers), we identified best practices for SE4ML, and composed a recommended reading list on the topic. These practices include the use of AutoML techniques to improve the use of ML components in software. All practices are described in our practice catalogue. We encourage readers to have a look and to send us their suggestions for additions.
Having a solid overview of the best practices, we set up a survey to find out the extent to which these SE4ML practices have already been adopted. We asked teams working on software with machine learning components to which degree they followed each practice in their work. In the resulting data we saw that tech companies have the highest adoption of the new ML related best practices, while research labs have the highest adoption of AutoML. Overall, more practices are adopted by teams that are larger and more experienced, with practices originating from TSE seeing slightly lower adoption than the new ML specific practices.
Effects of best practice adoption
Besides the adoption of best practices, we also investigated the effects of the practices. This resulted in insights into which specific practices relate to which desired outcomes. For instance, we found that software quality is influenced most by continuous integration, automated regression tests, and static analysis. On the other hand, agility is primarily affected by automated model deployment, having a cohesive- multi-disciplinary team, and enabling parallel training experiments. With these insights into the effectiveness of different practices, we hope to increase practice adoption and improve the quality of software with ML components.
A recent line of research advancements in ML focuses on automated machine learning (AutoML), an area that aims to automate (parts of) the construction and use of ML pipelines to enable a wider audience to make effective and responsible use of ML, without needing to become an expert in the field. We took a closer look at our survey results and found that, compared to non-tech companies and governmental organisations, research labs and tech companies are ahead in the adoption of AutoML practices (see Figure 2).
While overall AutoML adoption is similar across continents, non-profit and low-tech organisations see higher adoption in Europe than in North America. We also found that teams with multiple years of experience are more likely to adopt AutoML techniques. Finally, across the board, there is significant room to increase adoption of AutoML, but this is especially true for non-tech companies and governmental organisations.
Based on feedback from and interviews with participants, we recently revised our survey to learn more. Specifically, in our latest version of the survey, we ask in more detail about responsible use of ML and about the adoption of different AutoML techniques such as automated feature selection and neural architecture search. You can help us by taking the 10-minute survey and by spreading the word. If you want to stay up to date with our progress, keep an eye on our website.
Just two years after receiving thebest paper award at the 15th International Conference on Parallel Problem Solving from Nature (PPSN 2018), PhD student and Vanier Scholar Yasha Pushak and Professor Holger Hoos received anotherbest paper award at the 22nd international Genetic and Evolutionary Computation Conference (GECCO 2020) for a line of research on automated algorithm configuration. Their second paper, titled “Golden Parameter Search: Exploiting Structure to Quickly Configure Parameters in Parallel” was the single winner of the ECOM Track at GECCO 2020, determined by voting of the audience.
Important and challenging computation problems are typically solved through the use of highly parameterized algorithms. Finding high-quality parameter settings for the algorithm can often improve their performance by several orders of magnitude. In the last two decades, automated algorithm configuration has emerged as a hot topic of research.
State of the art algorithm configurators rely on powerful search heuristics to explore the space of parameter configurations; however, they are typically quite costly, often requiring days or even weeks to find high-quality parameter configurations. These methods assume that finding high-quality parameter configurations is challenging. Consider, for example, the problem of trying to climb to the highest point of a rugged mountain with a large number of peaks in a dense fog: If you always climb up the steepest hill, you will likely get stuck at a peak that is far from the highest on the mountain. To avoid this, most configurators spend time exploring the entire “landscapes” of the algorithm configuration problem — including regions that don’t initially appear promising, just to be certain they haven’t missed anything.
In their 2018 PPSNaward-winning paper, “Algorithm Configuration Landscapes: More Benign than Expected?” Yasha Pushak and Holger Hoos hypothesized that the landscapes induced by algorithm configuration problems are likely much simpler and more structured than previously assumed. By way of analogy, consider the task of choosing a temperature at which to bake a cake: If the temperature is too high the cake will burn, but if it is too low the cake won’t bake. Therefore, finding the optimal temperature simply corresponds to trading off between “too high” and “too low”. Yasha Pushak and Holger Hoos developed a statistical procedure to test for similar structure in the response of algorithm parameters. They applied their procedure to a large set of parameters and were unable to reject their hypotheses in 99.5% of the cases they studied. Read more about algorithm configuration landscapes here.
Capitalizing on these insights, Yasha Pushak and Holger Hoos developed a new automated algorithm configuration procedure: Golden Parameter Search (GPS). GPS is the first of a new generation of automated algorithm configuration procedures designed to exploit structural properties of algorithm configuration landscapes. In their 2020 GECCOaward-winning paper, they showed that GPS was often capable of finding similar-or better-quality parameter configurations using a fraction of the computational budget required by several other long-standing, state-of-the-art algorithm configuration procedures. Read more about GPS here.
Traffic congestion tends to be bad for the environment, the economy, and above all: the drivers’ moods. As such, it is a worthwhile cause to pursue improvements for; in particular, being computer- and data scientists, using data-driven methods to try to alleviate this problem seemed a particularly exciting approach. This is just what we (Laurens Arp, Dyon van Vreumingen, Daniela Gahwens and Mitra Baratchi) published a paper for in the IEEE MDM 2020 proceedings. The project, which started as a course project for Mitra’s Urban Computing course at Leiden University, evolved first into a short paper submission to the Netmob Future Cities Challenge (FCC), and subsequently into a full paper submission to the MDM conference.
The main idea of the method proposed was to redistribute traffic by imposing external costs (or rewards) to specific road segments. We used a movement dataset for Tokyo, provided by Foursquare for the FCC, from which we could derive which proportion of drivers would want to go to and from specific parts of the city. By combining this with a macro-scale traffic flow model (Greenshields), we were able to compute the occupancy of specific roads and the total travel time this resulted in. This model could then be used as an objective function for black-box optimisation; we would optimise the cost parameters of road segments so that the optimal number of desired routes got redirected to alternative roads such that the overall traffic time was minimised.
The amount of improvement (under the Greenshields model) we were able to achieve was highly dependent on the number of cars we estimated would be on the road at the same time. The best improvement achieved was 13.15% for a little over 13 500 cars (925 hours), and the worst was 1.35% for just under 113 000 cars (437 hours). Interestingly, even the relatively modest improvements, occurring for large amounts of cars, could still be meaningful, because there are more drivers to benefit from them. We also added a few fairness analyses to the paper, the results of which did not seem to indicate any unfair disadvantage to individual drivers. You can find the pre-recorded conference presentation here.
We hope that our paper will be able to contribute to the perpetually on-going efforts worldwide of causing less damage to the environment, boosting the economy, and perhaps also helping some drivers’ moods.
The ADA research group welcomes Bram Renting as a new PhD student.
Bram joined ADA in July 2020 as a PhD student under the supervision of Prof. dr. Holger Hoos and Prof. dr. Catholijn Jonker. He holds a master’s degree in computer science from TU Delft and has subsequently spent a year in the industry as a data scientist. He works at the ADA research group through the Hybrid Intelligence Centre consortium, which is a Dutch consortium that pushes for the development of human-AI collaboration.
Bram will research multi-agent systems where agents both have to cooperate and have individual preferences, so-called mixed-motive problems. His focus lies on scenarios where consensus between agents is required before actions are taken. Bargaining methods are used to reach this consensus. He starts with a focus on decentralised scheduling problems, which has applications in real life, such as calendar scheduling. The goal is to enhance human performance in these jobs through AI.
The ADA research group welcomes Matthias König as a new PhD student.
Before joining ADA, Matthias wrote his master thesis on automated age estimation from unconstrained facial imagery under the supervision of Holger Hoos and Jan van Rijn, while doing an internship at PwC’s Data Analytics unit. He holds a master’s degree in Media Technology from Leiden University and, next to that, followed the Information Studies/Data Science master’s course at the University of Amsterdam.
Matthias’ research is concerned with detecting when (Auto-)AI systems are “out of their depth” and developing mechanisms to fill potential gaps in the training space of these systems. Broadly speaking, he is interested in making AI systems more reliable by finding ways to verify their predictions – especially in settings where ground truth is inaccessible. Hopefully, his work will make AI-based tools safer to use for both experts and non-experts and reduce uncertainty in the predictions made by AI systems.
Koen van der Blom joined the ADA research group as a post-doctoral researcher. From March 2019 onward he started working with Holger Hoos in the area of meta-algorithmics. One of the things he works on is a tool called Sparkle. Sparkle aims to make meta-algorithmics such as algorithm selection and configuration easier to use for a wide audience. Besides this, he is also interested in performance analysis and prediction. What can be said about the expected performance of a new instance, based on previously seen instances? And how can you compare performance in a fair way?
Before this, during his PhD, Koen worked on multi-objective mixed-integer evolutionary algorithms applied to early-stage building design under the supervision of Michael Emmerich, Hèrm Hofmeyer, and Thomas Bäck. He continues to be interested in these problems, and particularly when it comes to optimisation in mixed-integer spaces. Who knows, perhaps combining aspects from the old and new will yet lead to other exciting work.
How could AI techniques be used to best improve the living standards of people around the world? This is the main question of interest for Laurens Arp, a Master student who joined the ADA research group in November 2019. He is currently working on his Master Thesis under the supervision of Mitra Baratchi and Holger Hoos. The project is about the data-driven evaluation and optimization methods of geographical regions.
The main focus of the project will be on (spatial) representation learning. Current methods would not sufficiently address the problem yet, as most approaches will either focus too much on spatial structure instead of how this structure affects the features of a neighborhood, or are aimed too much at encoding similarity rather than interaction. Once a suitable representation has been found, machine learning and deep learning could be used to automatically learn the relationship between geographical features and the measures one might like to use to evaluate a region. If the resulting model is sufficiently accurate, it could then be used to rate the quality of region configurations generated by an optimization algorithm, allowing for the optimization of the development of the region.