ERC FUNDED PROJECTS

Recurrent neural networks (RNNs) are general parallel-sequential computers. Some learn their programs or weights. Our supervised Long Short-Term Memory (LSTM) RNNs were the first to win pattern recognition contests, and recently enabled best known results in speech and handwriting recognition, machine translation, etc. They are now available to billions of users through the world's most valuable public companies including Google and Apple. Nevertheless, in lots of real-world tasks RNNs do not yet live up to their full potential. Although universal in theory, in practice they fail to learn important types of algorithms. This ERC project will go far beyond today's best RNNs through novel RNN-like systems that address some of the biggest open RNN problems and hottest RNN research topics: (1) How can RNNs learn to control (through internal spotlights of attention) separate large short-memory structures such as sub-networks with fast weights, to improve performance on many natural short-term memory-intensive tasks which are currently hard to learn by RNNs, such as answering detailed questions on recently observed videos? (2) How can such RNN-like systems metalearn entire learning algorithms that outperform the original learning algorithms? (3) How to achieve efficient transfer learning from one RNN-learned set of problem-solving programs to new RNN programs solving new tasks? In other words, how can one RNN-like system actively learn to exploit algorithmic information contained in the programs running on another? We will test our systems existing benchmarks, and create new, more challenging multi-task benchmarks. This will be supported by a rather cheap, GPU-based mini-brain for implementing large RNNs.

2 500 000 €

Start date: 2017-10-01, End date: 2022-09-30

"Recent technological evolutions, including the cloud, the multicore, the social and the mobiles ones, are turning computing ubiquitously distributed. Yet, building high-assurance distributed programs is notoriously challenging. One of the main reasons is that these systems usually seek to achieve several goals at the same time. In short, they need to be efficient, responding effectively in various average-case conditions, as well as reliable, behaving correctly in severe, worst-case conditions. As a consequence, they typically intermingle different strategies: each to cope with some specific condition, e.g., with or without node failures, message losses, time-outs, contention, cache misses, over-sizing, malicious attacks, etc. The resulting programs end up hard to design, prove, verify, implement, test and debug. Not surprisingly, there are anecdotal evidences of the fragility of the most celebrated distributed systems. The goal of this project is to contribute to building high-assurance distributed programs by introducing a new dimension for separating and isolating their concerns, as well as a new scheme for composing and reusing them in a modular manner. In short, the project will explore the inherent power and limitations of a novel paradigm, Adversary-Oriented Computing (AOC). Sub-programs, each implementing a specific strategy to cope with a given adversary, modelling a specific working condition, are designed, proved, verified, implemented, tested and debugged independently. They are then composed, possibly dynamically, as black-boxes within the same global program. The AOC project is ambitious and it seeks to fundamentally revisit the way distributed algorithms are designed and distributed systems are implemented. The gain expected in comparison with today's approaches is substantial, and I believe it will be proportional to the degree of difficulty of the distributed problem at hand."

2 147 012 €

Start date: 2014-06-01, End date: 2019-05-31

"Many problems in computer science can be cast as state-space search, where the objective is to find a path from an initial state to a goal state in a directed graph called a ""state space"". State-space search is challenging due to the state explosion problem a.k.a. ""curse of dimensionality"": interesting state spaces are often astronomically large, defying brute-force exploration. State-space search has been a core research problem in Artificial Intelligence since its early days and is alive as ever. Every year, a substantial fraction of research published at the ICAPS and SoCS conferences is concerned with state-space search, and the topic is very active at general AI conferences such as IJCAI and AAAI. Algorithms in the A* family, dating back to 1968, are still the go-to approach for state-space search. A* is a graph search algorithm whose only ""intelligence"" stems from a so-called ""heuristic function"", which estimates the distance from a state to the nearest goal state. The efficiency of A* depends on the accuracy of this estimate, and decades of research have pushed the envelope in devising increasingly accurate estimates. In this project, we question the ""A* + distance estimator"" paradigm and explore three new directions that go beyond the classical approach: 1. We propose a new paradigm of declarative heuristics, where heuristic information is not represented as distance estimates, but as properties of solutions amenable to introspection and general reasoning. 2. We suggest moving the burden of creativity away from the human expert by casting heuristic design as a meta-optimization problem that can be solved automatically. 3. We propose abandoning the idea of exploring sequential paths in state spaces, instead transforming state-space search into combinatorial optimization problems with no explicit sequencing aspect. We argue that the ""curse of sequentiality"" is as bad as the curse of dimensionality and must be addressed head-on."

1 997 510 €

Start date: 2019-02-01, End date: 2024-01-31

Autophagy is a fundamental cellular catabolic process deputed to the degradation and recycling of a variety of intracellular materials. Autophagy plays a significant role in multiple human physio-pathological processes and is now emerging as a critical regulator of skeletal development and homeostasis. We have discovered that during postnatal development in mice, the growth factor FGF18 induces autophagy in the chondrocyte cells of the growth plate to regulate the secretion of type II collagen, a major component of cartilaginous extracellular matrix. The FGF signaling pathways play crucial roles during skeletal development and maintenance and are deregulated in many skeletal disorders. Hence our findings may offer the unique opportunity to uncover new molecular mechanisms through which FGF pathways regulate skeletal development and maintenance and to identify new targets for the treatment of FGF-related skeletal disorders. In this grant application we propose to study the role played by the different FGF ligands and receptors on autophagy regulation and to investigate the physiological relevance of these findings in the context of skeletal growth, homeostasis and maintenance. We will also investigate the intracellular machinery that links FGF signalling pathways to the regulation of autophagy. In addition, we generated preliminary data showing an impairment of autophagy in chondrocyte models of Achondroplasia (ACH) and Thanathoporic dysplasia, two skeletal disorders caused by mutations in FGFR3. We propose to study the role of autophagy in the pathogenesis of FGFR3-related dwarfisms and explore the pharmacological modulation of autophagy as new therapeutic approach for achondroplasia. This application, which combines cell biology, mouse genetics and pharmacological approaches, has the potential to shed light on new mechanisms involved in organismal development and homeostasis, which could be targeted to treat bone and cartilage diseases.

1 586 430 €

Start date: 2017-01-01, End date: 2021-12-31