Report

Teaser, summary, work performed and final results

Periodic Reporting for period 1 - LAMBDA (Learning and Analysing Massive / Big complex Data)

Teaser

LAMBDA aims at transferring game changing technologies to the European industry in critical areas of Machine learning. Based on recent algorithmic breakthroughs, we adapt sophisticated methods to targeted industries to help turn cutting edge tools into innovative software...

Summary

LAMBDA aims at transferring game changing technologies to the European industry in critical areas of Machine learning. Based on recent algorithmic breakthroughs, we adapt sophisticated methods to targeted industries to help turn cutting edge tools into innovative software products and processes, tailored to real-world issues. LAMBDA focuses on two distinct application domains: 3D shape analysis and unstructured data mining. They share challenging features such as inherent complexity in modeling the data, high dimensionality which raises the issue of curse of dimensionality, and the need to address such datasets at a massive scale. 3D shape analysis is an important current problem in medicine, biology, as well as mechanical engineering and simulation. Given the huge success of manipulating speech (one dimensional) and images (2D), it is a natural next step to develop technology for 3D data. Our second domain of application is handling driving-insurance data so as to monitor driving behaviour, detect dangerous road segments, classify the latter in terms of accident frequency, and so on. For each application domain we include a significant industrial stakeholder within EU.

LAMBDA is characterised by a unique blend of theoretically rigorous and geometrically inclined methods, thus supporting a strong aspect of interdisciplinarity between Theory of Algorithms and Machine Learning. This shall be supported by advanced software development, ranging from public-domain prototype implementations to licensed software and integrated libraries. Our software methods are validated on synthetic data but also, when possible, on real datasets. LAMBDA strengthens existing links within Europe and across the Atlantic, while creating new synergies that support knowledge transfer beyond its lifetime.

Work performed

The project has achieved significant scientific progress as shown by the already published top-level research in relation to the 3 scientific Work-packages. In particular, research breakthroughs have been achieved on Dimensionality reduction by randomized linear projections, sampling and geometric random walks, Deep learning, Road segmentation / clustering, and risk assessment by anomaly detection. Four articles have been published in internationally established scientific journal or conferences with peer-reviewed proceedings, and one Poster has been presented by a secondee. Two workshops have been organized by the Consortium within the reporting period according to the provisions of the Grant Agreement.

The impact of the project can be summarized to the following points:
• Secondments have been undertaken between all members and partners. Career development of secondees (contacts with industry, experience of both sectors) has been significant. Transfer of Knowledge has been very satisfactory, both inter-sectoral and intercontinental, mainly by means of secondment visits but also participation in conferences.
• Communication is strong, given the current rate of publications as well as further activities such as use of LAMBDA to enhance course material and propose course projects, presentations in high-schools and University-level schools, general public talks. Some dissemination has occurred via LinkedIn and Twitter, as well as at events targeting the wider public (the coordinator has already participated in two general scientific conferences on Open Science, and on Math education).
• One industrial participant has showcased and exploited participation in LAMBDA in a recent funding round. An NDA was signed between an industrial and an academic member. Exploitable results elaborated and selected for EU\'s Innovation Radar concerning prospects for further innovation activities and transfer of technology in Shape representation, deep learning, Road segmentation / clustering, Driving behavior, anomaly detection, accident prediction. Joint software is being developed between academic and industrial members, and joint publications are in preparation.

Actions and deliverables relating to the LAMBDA project continue to take into account data privacy and confidentiality concerns, in order to ensure compliance with the relevant EU legislation. To that end, anonymisation and encryption techniques continue to be used, where appropriate, and specific privacy related information is not disclosed outside the context of the Project and save for its strictly defined purposes. Access restrictions to IT systems and storage such as password protection and ‘need-to-know’ access also continue to be in place, in accordance with the applicable legislation. All stakeholders inside the consortium are aware of the relevant data privacy and confidentiality concerns and requirements, notably those arising out of the General Data Protection Regulation (GDPR).

Final results

LAMBDA shall create a shared culture of research and innovation in crucial areas in Machine learning and Data Mining, and shall progress and innovate on a small number of critical applications in analysing complex data, namely 3D models, and road data aimed at the insurance and the financial business. A unique feature of LAMBDA is the mathematically rigorous geometric approach we bring into Machine Learning and Data Mining. Specifically, to address these challenges LAMBDA implements approximation algorithms. Our methods shall be validated on real-world data from our European industrial participants.

Technology transfer to and exploitation by the industrial participants as well as transfer of knowledge to the wider industrial community is our final goal. Specific results are expected in the following innovation aspects: First, efficient and compact representation of complex objects, including shapes with attributes, in high-dimensional space. Data structures and methods for efficient search and retrieval. New methods for matching and analysing 3D shapes. Second, clustering of complex data such as road segments. Analysis of car traffic and accident analysis leading to a publication and joint software. Both of these application domains are included in EU\'s Innovation Radar already.

An important aspect is to offer intersectoral training to all staff, especially young scientists, to create awareness of the role of businesses in technology, and the contribution of research organisations to innovation. We are providing international training at world leader institutions. Dissemination of LAMBDA’s results to the scientific community happens with publications and workshops. Communication of scientific and technological advances to the general public and in social media.

Website & more info

More info: http://lambda-project.eu/.