Transfer Learning to overcome catastrophic forgetting

In machine learning deep convolutional networks (deep learning) are very successful for solving particular problems [1] – at least when having many training samples. Great success has been made recently, e.g. in automatic game playing by AlphaGo (see the nature news here).  As fantastic these approaches are, it should be mentioned that deep learning has still serious limitations: they are black-box approaches, where it is currently difficult to explain how and why a result was achieved – see our recent work on glass-box approach [2], consequently lacking transparency and trust – issues which will become increasingly important in our data-centric world; they are demanding huge computational resources, and need enormous amounts of training data (often thousands, sometimes even millions of training samples) and standard approaches are poor at representing uncertainties which calls for Bayesian deep learning approaches [3]. Most of all, deep learning approaches are affected by an effect we call “catastrophic forgetting”.

What is catastrophic forgetting? One of the critical steps towards general artificial intelligence (human-level AI) is the ability to continually learn – similarly as we humans do: ongoing and continuous being capable of learning a new task B without forgetting how to perform an old task A. This seemingly trivial characteristic is not trivial for machine learning generally and deep learning specifically: McCloskey & Cohen already in 1989 [4] showed that neural networks have difficulties with this kind of transfer learning and coined the term catastrophic forgetting – and transfer learning is one attempt to overcoming it. Transfer learning is the ability to learn tasks permanently. Humans can do that very good – even very little children (refer to the work of Alison Gopnik, e.g. [5],  and at the bottom of this post). The synaptic consolidation in human brains may enable continual learning by reducing the plasticity of synapses that are vital to previously learned tasks ([6] and see a recent work on intelligent synapses for multi-task and transfer learning [7]). Based on these ideas the Google Deepmind Group around Demis Hassabis implemented a cool algorithm that performs a similar operation in artificial neural networks by constraining important parameters to stay close to their old values in their work on overcoming catastrophic forgetting in neural networks  (arXiv:1612.00796), [8]. As we know, a deep neural network consists of multiple layers of linear projections followed by element-wise non-linearities. Learning a task consists basically of adjusting the set of weights and biases θ of the linear projections, consequently, many configurations of θ will result in the same performance which is relevant for the so-called elastic weight consolidation (EWC): over-parametrization makes it likely that there is a solution for task B, θ ∗ B , that is close to the previously found solution for task A, θ ∗ A . While learning task B, EWC therefore protects the performance in task A by constraining the parameters to stay in a region of low error for task A centered around θ ∗ A. This constraint has been implemented as a quadratic penalty, and can therefore be imagined as a mechanical spring anchoring the parameters to the previous solution, hence the name elastic. In order to justify this choice of constraint and to define which weights are most important for a task, it is useful to consider neural network training from a probabilistic perspective. From this point of view, optimizing the parameters is tantamount to finding their most probable values given some data D. Interestingly, this can be computed as conditional probability p(θ|D) from the prior probability of the parameters p(θ) and the probability of the data p(D|θ) by: log p(θ|D) = log p(D|θ) + log p(θ) − log p(D).

Here exists and constantly emerge a lot of open and important research avenues which challenge the international machine learning community (see e.g. [9]).  The most interesting is what we don’t know yet – and the breaktrough machine learning approaches have not yet invented.

Andrew Y. Ng at the last NIPS 2016 conferene in Barcelona hold a tutorial where he emphasized the importance of transfer learning research and that “transfer learning will be the next driver of machine learning success” …
There is a wonderful post by Sebastian Ruder, see:

[1]          Yann Lecun, Yoshua Bengio & Geoffrey Hinton 2015. Deep learning. Nature, 521, (7553), 436-444, doi:10.1038/nature14539.

[2]          Andreas Holzinger, Markus Plass, Katharina Holzinger, Gloria Cerasela Crisan, Camelia-M. Pintea & Vasile Palade 2017. A glass-box interactive machine learning approach for solving NP-hard problems with the human-in-the-loop. arXiv:1708.01104.

[3]          Alex Kendall & Yarin Gal 2017. What Uncertainties Do We Need in Bayesian Deep Learning for Computer Vision? arXiv:1703.04977.

[4]          Michael Mccloskey & Neal J Cohen 1989. Catastrophic interference in connectionist networks: The sequential learning problem. In: Bower, G. H. (ed.) The Psychology of Learning and Motivation, Volume 24. San Diego (CA): Academic Press, pp. 109-164.

[5]          Alison Gopnik, Clark Glymour, David M Sobel, Laura E Schulz, Tamar Kushnir & David Danks 2004. A theory of causal learning in children: causal maps and Bayes nets. Psychological review, 111, (1), 3.

[6]          Stefano Fusi, Patrick J Drew & Larry F Abbott 2005. Cascade models of synaptically stored memories. Neuron, 45, (4), 599-611.

[7]          Friedemann Zenke, Ben Poole & Surya Ganguli. Continual Learning Through Synaptic Intelligence.  International Conference on Machine Learning, 2017. 3987-3995.

[8]          James Kirkpatrick, Razvan Pascanu, Neil Rabinowitz, Joel Veness, Guillaume Desjardins, Andrei A. Rusu, Kieran Milan, John Quan, Tiago Ramalho, Agnieszka Grabska-Barwinska, Demis Hassabis, Claudia Clopath, Dharshan Kumaran & Raia Hadsell 2017. Overcoming catastrophic forgetting in neural networks. Proceedings of the National Academy of Sciences, 114, (13), 3521-3526, doi:10.1073/pnas.1611835114.

[9]          Ian J Goodfellow, Mehdi Mirza, Da Xiao, Aaron Courville & Yoshua Bengio 2015. An empirical investigation of catastrophic forgeting in gradient-based neural networks. arXiv:1312.6211v3.



Machine learning researchers should watch the videos by Alison Gopnik, e.g.:

Machine Learning & Knowledge Extraction (MAKE) Journal launched

Inaugural Editorial Paper published:

Holzinger, A. 2017. Introduction to Machine Learning & Knowledge Extraction (MAKE). Machine Learning and Knowledge Extraction, 1, (1), 1-20, doi:10.3390/make1010001.

Machine Learning and Knowledge Extraction (MAKE) is an inter-disciplinary, cross-domain, peer-reviewed, scholarly open access journal to provide a platform to support the international machine learning community. It publishes original research articles, reviews, tutorials, research ideas, short notes and Special Issues that focus on machine learning and applications. Papers which deal with fundamental research questions to help reach a level of useable computational intelligence are very welcome.

Machine learning deals with understanding intelligence to design algorithms that can learn from data, gain knowledge from experience and improve their learning behaviour over time. The challenge is to extract relevant structural and/or temporal patterns (“knowledge”) from data, which is often hidden in high dimensional spaces,  thus not accessible to humans. Many application
domains, e.g., smart health, smart factory, etc. affect our daily life, e.g., recommender systems, speech recognition, autonomous driving, etc. The grand challenge is to understand the context in the real-world under uncertainty. Probabilistic inference can be of
great help here as the inverse probability allows to learn from data, to infer unknowns, and to make predictions to support decision making.

NOTE: To support the training of a new kind of machine learning graduates, the journal accepts peer-reviewed high-end tutorial papers, similar as the IEEE Signal Processing Magazine (SCI IF=9.654 !) is doing:

Call for Papers: Open Data for Discovery Science (due to July, 31, 2017)

The Journal BMC Medical Informatics and Decision Making (SCI IF (2015): 2,042)
invites to submit to a new thematic series on open data for discovery science

Note: Excellent submissions to the IFIP Cross Domain Conference on Machine Learning and Knowledge Discovery (CD-MAKE), (Submission due to May, 15, 2017) relevant to the topics described below, will be invited to expand their work into this thematic series:
The use of open data for discovery science has gained much attention recently as its full potential is unfolding and being explored in projects spanning all areas of healthcare research. A plethora of data sets are now available thanks to drives to make data universally accessible and usable for discovery science. However, with these advances come inherent challenges with the processing and management of ever expanding data sources. The computational and informatics tools and methods currently used in most investigational settings are often labor intensive and rely upon technologies that have not been designed to scale and support reasoning across multi-dimensional data resources. In addition, there are many challenges associated with the storage and responsible use of open data, particularly medical data, such as privacy, data protection, safety, information security and fair use of the data. There are therefore significant demands from the research community for the development of data management and analytic tools supporting heterogeneous analytic workflows and open data sources. Effective anonymisation tools are also of paramount importance to protect data security whilst preserving the usability of the data.

The purpose of this thematic series is to bring together articles reporting advances in the use of open data including the following:

  • The development of tools and methods targeting the reproducible and rigorous use of open data for discovery science, including but not limited to: syntactic and semantic standards, platforms for data sharing and discovery, and computational workflow orchestration technologies that enable the creation of data analytics, machine learning and knowledge extraction pipelines.
  • Practical approaches for the automated and/or semi-automated harmonization, integration, analysis, and presentation of data products to enable hypothesis discovery or testing.
  • Theoretical and practical approaches for solutions to make use of interactive machine learning to put a human-in-the-loop, answering questions including: could human intelligence lead to general heuristics that we can use to improve heuristics?
  • Frameworks for the application of open data in hypothesis generation and testing in projects spanning translational, clinical, and population health research.
  • Applied studies that demonstrate the value of using open data either as a primary or as an enriching source of information for the purposes of hypothesis generation/testing or for data-driven decision making in the research, clinical, and/or population health environments.
  • Privacy preserving machine learning and knowledge extraction algorithms that can enable the sharing of previously “privileged” data types as open data.
  • Evaluation and benchmarking methodologies, methods and tools that can be used to demonstrate the impact of results generated through the primary or secondary use of open data.
  • Socio-cultural, usability, acceptance, ethical and policy issues and frameworks relevant to the sharing, use, and dissemination of information and knowledge derived from the analysis of open data.

Submission is open to everyone, and all submitted manuscripts will be peer-reviewed through the standard BMC Medical Informatics and Decision Making review process. Manuscripts should be formatted according to the submission guidelines and submitted via the online submission system. Please indicate clearly in the covering letter that the manuscript is to be considered for the ‘Open data for discovery science’ collection. The deadline for submissions will be 31 July 2017.

For further information, please email the editors of the thematic series:
Philip PAYNE ,or the BMC in-house editor

Link to the IFIP Cross-Domain Conference on Machine Learning and Knowledge Extraction (CD-MAKE):

Federated Collaborative Machine Learning

The Google Research Group [1] is always doing awesome stuff, the most recent one is on Federated Learning [2], which enables e.g. smart phones (of course any computational device, and maybe later all internet-of-things, intelligent sensors in either smart hospitals or in smart factories etc.) to collaboratively learn a shared representation model, whilst keeping all the training data on the local devices, decoupling the ability to do machine learning from the need to store the data centralized in the cloud. This goes beyond the use of local models that make predictions on mobile devices (like the Mobile Vision API and On-Device Smart Reply) by bringing model training to the device as well – which is great. The problem with standard approaches is that you always need centralized training data – either on your USB-stick, as the medical doctors do, or in a sophisticated centralized data center.

The basic idea is that the mobile device downloads the current modela and subsequently improves it by learning from data on the respective device, and then summarizes the changes as a small focused update. The remarkable detail is that only this update to the model is sent to the cloud (yes, here privacy, data protection safety and security is challenged see e.g. [3] – but this is much easier to do with this small data – as when you would do it with the raw data – think for example on patient data), where it is immediately averaged with other devicer updates to improve the shared model. All the training data remains on the local devices, and no individual updates are stored in the cloud.

The Google Group recently solved a lot of algorithmic and technical challenges. In a typical machine learning system, an optimization algorithm e.g. Stochastic Gradient Descent (SGD) [4] runs on a large dataset partitioned homogeneously across servers in the cloud. Such highly iterative algorithms require low-latency, high-throughput connections to the training data. But in the Federated Learning setting, the data is distributed across millions of devices in a highly uneven fashion. In addition, these devices have significantly higher-latency, lower-throughput connections and are only intermittently available for training.

This calls for a lot of further investigations with interactive Machine Learning (iML) bringing the human-into-the loop, i.e. making use of human cognitive abilities. This can be of particular interest to solve problems, where learning algorithms suffer due to insufficient training samples (rare events, single events), where we deal with complex data and/or computationally hard problems. For example, “doctors-in-the-loop” can help with their long-term experience and heursitic knowledge to solve problems which otherwise would remain NP-hard [5, 6]. A further step is with many humans-in-the-loop: Such collaborative interactive Machine Learning (ciML) can help in many application areas and domains, e.g. in in health informatics (smart hospital) or in industrial applications (smart factory) [7].

Read the original article, posted on April, 6, 2017,  here:


[2] NIPS Workshop on Private Multi-Party Machine Learning, Barcelona, December, 9, 2016,

[3] Bonawitz, K., Ivanov, V., Kreuter, B., Marcedone, A., Mcmahan, H. B., Patel, S., Ramage, D., Segal, A. & Seth, K. 2016. Practical Secure Aggregation for Federated Learning on User-Held Data. arXiv preprint arXiv:1611.04482.

[4] Bottou, L. 2010. Large-scale machine learning with stochastic gradient descent. Proceedings of COMPSTAT’2010. Springer, pp. 177-186. doi:10.1007/978-3-7908-2604-3_16  (N.B.: 836 citations as of 08.04.2017)

[5] Holzinger, A. 2016. Interactive Machine Learning for Health Informatics: When do we need the human-in-the-loop? Brain Informatics, 3, (2), 119-131, doi:10.1007/s40708-016-0042-6

[6] Holzinger, A., Plass, M., Holzinger, K., Crisan, G., Pintea, C. & Palade, V. 2016. Towards interactive Machine Learning (iML): Applying Ant Colony Algorithms to solve the Traveling Salesman Problem with the Human-in-the-Loop approach. In: Springer Lecture Notes in Computer Science LNCS 9817. Heidelberg, Berlin, New York: Springer, pp. 81-95, [pdf]

[7] Robert, S., Büttner, S., Röcker, C. & Holzinger, A. 2016. Reasoning Under Uncertainty: Towards Collaborative Interactive Machine Learning. In: Machine Learning for Health Informatics: Lecture Notes in Artifical Intelligence LNAI 9605. Springer, pp. 357-376, [pdf]

Image source:


Integrated interactomes and pathways in precision medicine by Igor Jurisica, Toronto

Machine learning is the fastest growing field in computer science, and Health Informatics is amongst the greatest application challenges, providing benefits in improved medical diagnoses, disease analyses, and pharmaceutical development – towards future precision medicine.

Talk announcement: Friday, 12th May, 2017, 10:00, Seminaraum 137, Parterre, Inffeldgasse 16c

Integrated interactomes and pathways in precision medicine

by Igor Jurisica, University of Toronto and Princess Margaret Cancer Center Toronto

Abstract: Fathoming cancer and other complex disease development processes requires systematically integrating diverse types of information, including multiple high-throughput datasets and diverse annotations. This comprehensive and integrative analysis will lead to data-driven precision medicine, and in turn will help us to develop new hypotheses, and answer complex questions such as what factors cause disease; which patients are at high risk; will patients respond to a given treatment; how to rationally select a combination therapy to individual patient, etc.
Thousands of potentially important proteins remain poorly characterized. Computational biology methods, including machine learning, knowledge extraction, data mining and visualization, can help to fill this gap with accurate predictions, making disease modeling more comprehensive. Intertwining computational prediction and modeling with biological experiments will lead to more useful findings faster and more economically.

Short Bio: Igor Jurisica is Tier I Canada Research Chair in Integrative Cancer Informatics, Senior Scientist at Princess Margaret Cancer Centre, Professor at University of Toronto and Visiting Scientist at IBM CAS. He is also an Adjunct Professor at the School of Computing, Pathology and Molecular Medicine at Queen’s University, Computer Science at York University, scientist at the Institute of Neuroimmunology, Slovak Academy of Sciences and an Honorary Professor at Shanghai Jiao Tong University in China. Since 2015, he has also served as Chief Scientist at the Creative Destruction Lab, Rotman School of Management. Igor has published extensively on data mining, visualization and cancer informatics, including multiple papers in Science, Nature, Nature Medicine, Nature Methods, Journal of Clinical Oncology, and received over 9,960 citations since 2012. He has been included in Thomson Reuters 2016, 2015 & 2014 list of Highly Cited Researchers, and The World’s Most Influential Scientific Minds: 2015 & 2014 Reports.

Jurisica Lab, IBM Life Sciences Discovery Center:

Canada Tier I Research Chair:

On Nutrigenomics [1]:

[1] Nutrigenomics tries to define the causality or relationship between specific nutrients and specific nutrient regimes (diets) on human health. The underlying idea is in personalized nutrition based on the *omics background, which may help to foster personal dietrary recommendations. Ultimately, nutrigenomics will allow effective dietary-intervention strategies to recover normal homeostasis and to prevent diet-related diseases, see: Muller, M. & Kersten, S. 2003. Nutrigenomics: goals and strategies. Nature Reviews Genetics, 4, (4), 315-322.

What is machine learning?

Many services of our every day life rely meanwhile on machine learning – a field of science and a powerful technology that allows machines to learn from data; a very nice info graphic by the Royal Society – interactive with a quiz – can be found here:

Royal Society Infographic “What is machine learning?”

This is part of a info campaign about machine learning from the Royal Society:

The Royal Society was formed by a group of natural scientists influenced by Francis Bacon (1561-1626).  The first ‘learned society’ meeting on 28 November 1660 followed a lecture at Gresham College by Christopher Wren. Joined by Robert Boyle and John Wilkins and others, the group received royal approval by King Charles II (1630-1685) in 1663 and was known since as ‘The Royal Society of London for Improving Natural Knowledge’.

Machine Learning Guide

An excellent podcast which I can fully recommend to my students is the Machine Learning Guide by Tyler RENELLE (Tensor Flow). This series aims to teach the high level fundamentals of machine learning with a focus on algorithms and some underlying mathematics, which is really great.




CD-MAKE machine learning and knowledge extraction

Cross Domain Conference for Machine Learning & Knowledge Extraction

Call for Papers – due to May, 15, 2017

Call for Papers due to May, 15, 2017

International IFIP Cross Domain Conference for Machine Learning & Knowledge Extraction CD-MAKE
in Reggio di Calabria (Italy) August 29 – September 1, 2017

CD stands for Cross-Domain and means the integration and appraisal of different fields and application domains (e.g. Health, Industry 4.0, etc.) to provide an atmosphere to foster different perspectives and opinions. The conference is dedicated to offer an international platform for novel ideas and a fresh look on the methodologies to put crazy ideas into Business for the benefit of the human. Serendipity is a desired effect, and shall cross-fertilize methodologies and transfer of algorithmic developments.

MAKE stands for MAchine Learning & Knowledge Extraction.

CD-MAKE is a joint effort of IFIP TC 5, IFIP WG 8.4, IFIP WG 8.9 and IFIP WG 12.9 and is held in conjunction with the International Conference on Availability, Reliability and Security (ARES).
Keynote Speakers are Neil D. LAWRENCE (Amazon) and Marta MILO (University of Sheffield).

IFIP is the International Federation for Information Processing and the leading multi-national, non-governmental, apolitical organization in Information & Communications Technologies and Computer Sciences, is recognized by the United Nations and was established in the year 1960 under the auspices of the UNESCO as an outcome of the first World Computer Congress held in Paris in 1959.

Papers are sought from the following seven topical areas (see image below). Papers which deal with fundamental questions and theoretical aspects in machine learning are very welcome.

❶ Data science (data fusion, preprocessing, data mapping, knowledge representation),
❷ Machine learning (both automatic ML and interactive ML with the human-in-the-loop),
❸ Graphs/network science (i.e. graph-based data mining),
❹ Topological data analysis (i.e. topology data mining),
❺ Time/entropy (i.e. entropy-based data mining),
❻ Data visualization (i.e. visual analytics), and last but not least
❼ Privacy, data protection, safety and security (i.e. privacy aware machine learning).

Proposals for Workshops, Special Sessions, Tutorials: April, 19, 2017
Submission Deadline: May, 15, 2017
Author Notification: June, 14, 2017
Camera Ready Deadline: July, 07, 2017


Stan: A probabilistic programming language

A long time ago submitted paper from the Stan developers
has finally been appeared at the Journal of statistical software:

Carpenter, B., Gelman, A., Hoffman, M., Lee, D., Goodrich, B., Betancourt, M., Brubaker, M. A., Guo, J., Li, P. & Riddell, A. 2017. Stan: A probabilistic programming language. Journal of Statistical Software, 76, (1), 1-32, doi:10.18637/jss.v076.i01

Also the Python package can be downloaded from the site!

Stan is a probabilistic programming language for specifying statistical models. A Stan program imperatively defines a log probability function over parameters conditioned on specified data and constants. Stan provides full Bayesian inference
for continuous-variable models through Markov chain Monte Carlo methods such as the No-U-Turn sampler, an adaptive form of Hamiltonian Monte Carlo sampling. Penalized maximum likelihood estimates are calculated using optimization methods such as the limited memory Broyden-Fletcher-Goldfarb-Shanno algorithm. Stan is also a platform for computing log densities and their gradients and Hessians, which can be used in alternative algorithms such as variational Bayes, expectation propagation, and marginal inference using approximate integration. To this end, Stan is set up so that the densities, gradients, and Hessians, along with intermediate quantities of the algorithm such as acceptance probabilities, are easily accessible. Stan can be called from the command line using the cmdstan package, through R using the rstan package, and through Python using the pystan package. All three interfaces support sampling and optimization-based inference with diagnostics and posterior analysis. rstan and pystan also provide access to log probabilities, gradients, Hessians, parameter transforms, and specialized plotting.

Congratulations from the Holzinger Group to the authors!

Machine Learning Podcast: Data Skeptic (recommendable)

Data Skeptic is a weekly podcast that is skeptical of and with data. They explain methods and algorithms that power our world in an accessible manner through short mini-episode discussions and longer interviews with experts in the field, see: