Search

In Materio computation aims to leverage the complex physics that dictates the response of metamaterials to perform computational operations. Magnetic materials are of particular interest for In Materio computation due to their complex nonlinear interactions to stimuli, non-volatility of response, and ease of electrical interfacing with existing computational platforms. Magnetic materials can also exhibit emergent behaviours -- where local interactions lead to complex global behaviours that are robust to variation in individual elements -- which are promising for In Materio computation due to the high degree of expressivity that emergence offers. However, to make use of the broad range of complexity that these emergent systems boast, methods for reading and writing data to and from the material substrate must be tailored to extract the specific computational properties that a given task demands. Here, devices of arrays of interconnected permalloy nanorings, shown to exhibit emergent dynamics, are electrically measured under three input configurations falling under the paradigm of 'Reservoir Computing' to perform signal transformation, spoken digit recognition, and memory-intensive time series prediction. The successful performance of each of these tasks is then correlated to the physical processes occurring in the device, providing interpretability of the role of the substrate in computation. The ring array system performs competitively compared to other proposed magnetic architectures in the signal transformation and spoken digit recognition tasks, and provides, to the author's knowledge, the first experimental demonstration of NARMA time series prediction in a nanomagnetic platform without the use of external delay lines. This constitutes a critical step forward in the design of In Materio computation devices which can exploit the rich dynamics offered by emergent systems.

Neural networks have revolutionized the area of artificial intelligence and introduced transformative applications to almost every scientific field and industry. However, this success comes at a great price; the energy requirements for training advanced models are unsustainable. One promising way to address this pressing issue is by developing low-energy neuromorphic hardware that directly supports the algorithm's requirements. The intrinsic non-volatility, non-linearity, and memory of spintronic devices make them appealing candidates for neuromorphic devices. Here, we focus on the reservoir computing paradigm, a recurrent network with a simple training algorithm suitable for computation with spintronic devices since they can provide the properties of non-linearity and memory. We review technologies and methods for developing neuromorphic spintronic devices and conclude with critical open issues to address before such devices become widely used.

Neuromorphic computing leverages the complex dynamics of physical systems for computation. The field has recently undergone an explosion in the range and sophistication of implementations, with rapidly improving performance. Neuromorphic schemes typically employ a single physical system, limiting the dimensionality and range of available dynamics - restricting strong performance to a few specific tasks. This is a critical roadblock facing the field, inhibiting the power and versatility of neuromorphic schemes. Here, we present a solution. We engineer a diverse suite of nanomagnetic arrays and show how tuning microstate space and geometry enables a broad range of dynamics and computing performance. We interconnect arrays in parallel, series and multilayered neural network architectures, where each network node is a distinct physical system. This networked approach grants extremely high dimensionality and enriched dynamics enabling meta-learning to be implemented on small training sets and exhibiting strong performance across a broad taskset. We showcase network performance via few-shot learning, rapidly adapting on-the-fly to previously unseen tasks.

We examined how bees solve a visual discrimination task that uses stimuli commonly found in numerical cognition studies. Bees performed well on the task, but additional tests showed that they had learned continuous (non-numerical) cues. A network model using biologically plausible visual feature filtering and a simple associative rule was capable of learning the task using only continuous cues inherent in the training stimuli, with no numerical processing. This model was also able to reproduce behaviours that have been considered in other studies indicative of numerical cognition. Our results support the idea that a sense of magnitude may be more primitive and basic than a sense of number. Our findings highlight how problematic inadvertent continuous cues can be for studies of numerical cognition. This remains a deep issue within the field that requires increased vigilance and cleverness from the experimenter. We suggest ways of better assessing numerical cognition in non-speaking animals, including assessing the use of all alternative cues in one test, using cross-modal cues, analysing behavioural responses to detect underlying strategies and finding the neural substrate.

Eνήλικα ή σωματικά βλαστικά κύτταρα ορίζονται τα κύτταρα με δυνατότητα ανανέωσης και διαφοροποίησης σε όλους τους κυτταρικούς τύπους ενός ιστού. Είναι σημαντικά για τη φυσιολογική ανανέωση του ιστού και ενεργοποιούνται κατόπιν τραυματισμού του. Σε συγκεκριμένους κυτταρικούς τύπους έχει βρεθεί ότι αποτελούν τα κύτταρα προέλευσης του καρκίνου. Ένα μονοπάτι που εμπλέκεται τόσο στη διατήρηση των βλαστικών κυττάρων όσο και στη διαφοροποίησή τους είναι το μονοπάτι σηματοδότησης Notch. Αποτελεί ένα εξελικτικά συντηρημένο μονοπάτι σε όλα τα μετάζωα, που εμπλέκεται στην εμβρυϊκή ανάπτυξη και την ομοιόσταση των ιστών. Κατά συνέπεια, η δυσλειτουργία ή η απώλεια αυτού του μονοπατιού οδηγεί σε πληθώρα ανθρώπινων ασθενειών, όπως ο καρκίνος. Οι λειτουργίες του στηρίζονται κυρίως στη διασύνδεση γειτονικών κυττάρων που επιτυγχάνεται μέσω της φυσικής επαφής των διαμεμβρανικών υποδοχέων Notch (1-4) και των ομοειδών προσδετών αυτών. Στα θηλαστικά, υπάρχουν 5 προσδέτες: Dll1 (Delta-like 1), Dll3, Dll4, Jagged1 και Jagged2. Η συγκεκριμένη μελέτη επικεντρώνεται στην αξία του Dll1 ως κρίσιμου δείκτη βλαστικών κυττάρων. Πρωταρχικός σκοπός της παρούσας μελέτης είναι η εύρεση επιπρόσθετων δεικτών βλαστικών/προγονικών κυττάρων σε ποικιλία ιστών του ποντικού. Με στόχο τη διερεύνηση του ρόλου του προσδέτη του μονοπατιού Notch, Dll1, στην ανάπτυξη του ποντικού, δημιουργήσαμε ένα διαγονιδιακό «knock-out/knock-in» ποντικό που εκφράζει τη ρεκομπινάση Cre υπό τον έλεγχο του υποκινητή του γονιδίου Dll1. Αξιοποιήσαμε μία επαγόμενη μορφή ρεκομπινάσης, η οποία ενεργοποιείται μόνο μετά από ένεση με Tamoxifen. Με τη βοήθεια του ποντικού Dll1CreERT2, έχουμε χρησιμοποιήσει in vivo και in vitro προσεγγίσεις για να σημάνουμε γενετικά σπάνιους υποπληθυσμούς κυττάρων που εκφράζουν το Dll1 σε διάφορους ιστούς του ενήλικου ποντικού, με έμφαση στο μαστικό αδένα, αλλά και το στόμαχο και το δέρμα, που βρέθηκαν να ανταποκρίνονται στην ογκογόνο σηματοδότηση του KRAS. Προχωρήσαμε σε διασταυρώσεις του ποντικού Dll1CreERT2 με στελέχη επαγόμενης έκφρασης γονιδίου-μάρτυρα. Χρησιμοποιώντας τους διπλά διαγονιδιακούς ποντικούς Dll1CreERT2;Rosa26mTmG, ανιχνεύσαμε Dll1-θετικά κύτταρα σε πολλούς ιστούς, όπως στο μαστικό αδένα, το πάγκρεας, το έντερο, το στόμαχο, το νεφρό, την ουροδόχο κύστη, το ήπαρ, τον σπλήνα και το δέρμα. Η τοπολογία των Dll1θετικών κυττάρων σε πολλούς από αυτούς τους ιστούς είναι συμβατή με τη βιβλιογραφία, ως προς τη θέση που έχει βρεθεί να εδράζονται τα βλαστικά κύτταρα. Για τη διερεύνηση της ανταπόκρισης των Dll1-θετικών κυττάρων σε ογκογόνες μεταλλαγές και την τυχόν συμμετοχή τους ως κύτταρα προέλευσης του καρκίνου, εκφράσαμε μεταλλαγμένα ογκογονίδια ή ογκοκατασταλτικά γονίδια υπό τον έλεγχο του Dll1 υποκινητή. Εντυπωσιακό εύρημα αυτής της σειράς πειραμάτων ήταν οι νεοπλασίες που δημιουργήθηκαν στο δέρμα και τον στόμαχο σε ελάχιστο χρονικό διάστημα από την ένεση με Tamoxifen, δηλαδή από τη στιγμή που εκφράστηκε η μεταλλαγμένη μορφή KRAS(G12D) στα κύτταρα Dll1. Ενεργές μεταλλαγές στο KRAS είναι οι πιο συχνές μεταλλαγές στον καρκίνο του ανθρώπου, κυρίως σε επιθηλιακούς ιστούς. Ιστολογική ανάλυση των νεοπλασιών στον στόμαχο και το δέρμα υπέδειξε πλακώδη υπερπλασία, αλλά όχι διήθηση. Συμπερασματικά, τα σπάνια κύτταρα Dll1 στο πλακώδες επιθήλιο του προστόμαχου, καθώς και του δέρματος, αντιπροσωπεύουν προγονικά κύτταρα που είναι ευαίσθητα στην ογκογόνο δράση του KRAS. Στον στόμαχο, το μοντέλο αυτό αποτελεί ισοδύναμο σύστημα για τον καρκίνο Barrett’s του οισοφάγου του ανθρώπου. Επίσης, στους ποντικούς Dll1CreERT2;KRASG12D παρατηρήθηκαν έντονες υπερπλασίες, είτε απλές είτε άτυπες, στα κύτταρα αυλού των αγωγών του μαστικού αδένα, ένα προκαρκινικό στάδιο του μαστού. Αυτή η παρατήρηση, σε τόσο σύντομο χρονικό διάστημα, συνεισφέρει σημαντικά στο παραπάνω συμπέρασμα, ότι δηλαδή τα κύτταρα Dll1+ είναι τα προγονικά κύτταρα αυλού και αποτελούν τα κύτταρα προέλευσης του καρκίνου του μαστού. Τα κύτταρα που εκφράζουν Dll1 παρουσιάστηκαν ευαίσθητα και στις περιπτώσεις που υπερεκφράσαμε το KRAS, αλλά και τη μεταλλαγμένη μορφή PIK3CA(H1047R), μόνο στα Dll1-θετικά κύτταρα. Το γονίδιο PIK3CA είναι το πιο συχνά μεταλλαγμένο γονίδιο στον καρκίνο του μαστού στον άνθρωπο, με εξαίρεση φυσικά το TP53. Πραγματοποιώντας δοκιμασίες Fat Pad (ορθόλογες μεταμοσχεύσεις στο μαστικό αδένα), πειράματα γενεαλογικού εντοπισμού in vivo στο μαστικό αδένα και κλωνογενετικά πειράματα in vitro, συμπεράναμε ότι το Dll1 εκφράζεται τόσο στα μυοεπιθηλιακά κύτταρα όσο και στα κύτταρα αυλού των αγωγών του μαστού. Πιστεύουμε στην ύπαρξη διδύναμων βλαστικών κυττάρων μόνο κατά την εμβρυογένεση, τα οποία εξαφανίζονται κατά την ωρίμανση του μαστικού αδένα και αντικαθίστανται από προγονικά κύτταρα, ειδικά για την κάθε κυτταρική στοιβάδα, που σημαίνονται από το Dll1. Τα κύτταρα αυτά αποτελούν τα κύτταρα που πολλαπλασιάζονται και επεκτείνονται κατά την ανάπτυξη του αδένα, αλλά και κατά τις εγκυμοσύνες. Επίσης, η στατιστικά σημαντική αύξηση των Dll1-θετικών κυττάρων αυλού μετά από πολλαπλές εγκυμοσύνες, μας δείχνει ότι αυτά τα κύτταρα συμβάλλουν στην αναδιαμόρφωση του ιστού μετά από απογαλακτισμό και υποστροφές του. Η στοχευμένη απαλοιφή των Dll1-θετικών κυττάρων οδηγεί σε αδυναμία της κλωνογενετικής ικανότητας του μαστικού αδένα. Μόνο τα Dll1-θετικά μυοεπιθηλιακά κύτταρα είναι ικανά να ανασυστήσουν έναν πλήρως λειτουργικό μαστικό αδένα κατά τις δοκιμασίες Fat Pad. Τέλος, η υπερέκφραση του hN1IC (human Notch 1 Intracellular Domain) στα κύτταρα Dll1 οδηγεί σε πολυκυστική κατάσταση του ήπατος (προκαρκινικό στάδιο πριν το ενδοηπατικό χολαγγειοκαρκίνωμα) μέσα σε διάστημα 6-12 μηνών. Εκκρεμούν πειράματα γενεαλογικού εντοπισμού, καθώς και αναγέννησης του ήπατος, για να ελέγξουμε το ρόλο των κυττάρων Dll1 στο ήπαρ. Στην παρούσα Διδακτορική Διατριβή, καταλήγουμε στο συμπέρασμα ότι τα Dll1-θετικά κύτταρα αποτελούν βλαστικά/προγονικά κύτταρα σε εύρος ιστών του ποντικού, όπως στον στόμαχο, το δέρμα και το μαστικό αδένα. Κατ’ επέκταση, αυτά είναι τα κύτταρα που ανταποκρίνονται σε ογκογόνες μεταλλαγές και αποτελούν τα κύτταρα προέλευσης του καρκίνου στους συγκεκριμένους ιστούς. Besides embryonic, there are the adult or somatic stem cells. By definition, they have the ability to self-renew and differentiate into all cell types of a tissue. They are important for the natural tissue regeneration and can be mobilized upon injury. For certain cell types, it has been shown that they are the cells of origin of cancer. To isolate adult stem cells, there are markers based on functional methods. For validation of them, there are in vitro clonogenic assays and in vivo transplantations. One of the pathways involved both in stem cell maintenance and differentiation is the Notch signaling pathway. It is evolutionarily conserved in all metazoans, and plays a crucial role in embryonic development and tissue homeostasis. Therefore, deregulation or loss of Notch signaling leads to a wide spectrum of human diseases, including cancer, in various tissues. Its functions rely primarily upon the crosstalk of neighboring cells achieved through the physical interaction of transmembrane receptors (Notch 1-4) and cognate ligands. In mammals, there are 5 ligands: Dll1 (Delta-like 1), Dll3, Dll4, Jagged1 and Jagged2. This study focuses on the value of Dll1 as a putative stem/progenitor cell marker. The aim of the present study is to find additional markers of stem/progenitor cells in a variety of mouse tissues. In order to gain insight in the role of the Notch ligand Dll1 in mouse development, we generated a “knock-out/knock-in” line expressing a Cre recombinase. We use an inducible form of Cre recombinase (CreER), which becomes active only upon intraperitoneal injection with Tamoxifen. We have employed in vivo and in vitro approaches in mice to genetically mark rare subpopulations of cells expressing Dll1 in various adult tissues, with emphasis on the mammary gland, and stomach and skin that responded to KRAS oncogenic events. To assess recombinase activity, we crossed Dll1CreERT2 mouse with inducible fluorescent reporter lines, like the Rosa26mTmG mouse strain, which constitutively expresses a membrane-anchored tomato fluorescent protein and switches to membranous GFP expression upon Cre recombination. Using Dll1CreERT2;Rosa26mTmG bitransgenic mice, we observed Dll1-positive cells in the mammary gland, pancreas, intestine, stomach, kidney, bladder, liver, spleen and skin. In some of these tissues, the topology of Dll1 cells is compatible with literature, as to where stem cells are thought to reside. Activating mutations of KRAS are found in high percentage in various, epithelial mostly, tissues. Among the mutations of KRAS, the KRAS(G12D) is the most frequent one. To investigate the responsiveness of Dll1-positive cells to KRAS oncogenic signaling and their possible involvement as cells of origin of cancer in various tissues, we crossed KRAS(G12D) mice with Dll1CreERT2. An impressive finding was that when we conditionally expressed a constitutively active RAS oncoprotein in Dll1-positive cells, mice developed neoplasias in the skin, as well as in the stomach, only within days. Histological analysis of these neoplasms indicated squamous hyperplasias, but not invasion. So, the sparse Dll1 cells in the squamous forestomach and skin represent progenitor cells that are sensitive to KRAS oncogenic signaling. In the stomach, this is an equivalent model for human Barrett’s esophagus cancer. In addition, we expressed oncogenes or tumor suppressor genes, under the control of Dll1 and studied the effect of these genetic mutations on the Dll1 transformation process in the mammary gland. In Dll1CreERT2;KRAS(G12D) mice, we observed vigorous hyperplasia (simple, atypical or in situ) of luminal cells in mammary ducts, which is a precancerous stage in breast. This observation, in such a short period of time, contributes to the aforementioned conclusion that Dll1 cells are the progenitor cells of the luminal cells, from which breast cancer originates. Probably, due to the mortality of these mice, because of stomach and lips’ tumors, we were unable to observe tumors in the mammary gland. When we overexpressed KRAS in Dll1-positive cells, we obtained a more aggressive phenotype. Finally, PIK3CA is the most commonly mutated gene in human breast cancer, with the exception of TP53. When the most frequent mutation PIK3CA(H1047R) was expressed exclusively in Dll1 cells, we got similar hyperplasias of the ducts in the mammary gland. With Fat Pad assays, lineage tracing in vivo experiments in mammary gland and clonogenic assays in vitro, we concluded that Dll1 is expressed in both myoepithelial and luminal cells of mammary ducts. We believe that there are bipotent progenitors only during embryogenesis and they disappear during maturation of the mammary gland. Then, committed-lineage progenitors that are labeled by Dll1 are responsible for the regeneration of mammary gland. The increase of the number of Dll1+ luminal cells after multiple pregnancies implies that these cells contribute to mammary reconstraction following mouse weaning and mammary gland involution. Evadication of Dll1 cells has as a result the elimination of clonogenic capacity of mammary gland. Dll1-expressing cells are sensitive to KRAS signaling. Only Dll1 positive myoepithelial cells can reconstitute a fully functional mammary gland, under limiting dilutions. Finally, overexpression of hN1IC in Dll1 cells leads to a polycystic state of the liver, within 6 months to 1 year, a state before intrahepatic cholangiocarcinoma. In our future plans, we shall conduct experiments of lineage tracing and regeneration of liver, in order to prove that Dll1 cells are stem cells in the liver. In this PhD Thesis, we conclude that Dll1 cells are stem/progenitor cells in various mouse tissues, apart from intestine, where it is already known that these cells are the stem cells upon damage of the tissue, and that cancer originates from these special type cells.

This talk aims describes work in the intersection of Computational Neuroscience and Machine Learning as well as difficulties and failures of research. The work presented was published as: Vasilaki, E. (2017) Is Epicurusthe father of Reinforcement Learning, arXiv:1710.04582v1 Richmond, P., Buesing, L., Giugliano, M., & Vasilaki, E. (2011), Democratic population decisions result in robust policy-gradient learning: a parametric study with GPU simulations. PLOS One, 6(5):e18539. doi:10.1371/journal.pone.0018539. Vasilaki, E., Fremaux, N.,Urbanczik, R., Senn, W, and Gerstner, W. (2009), Spike-based reinforcement learning in continuous State and Action Space: when policy gradient methods fail. PLOS Computational Biology, Vol.5(12):e1000586 doi:10.1371/journal.pcbi.1000586.

The insect central complex (CX) is an enigmatic structure whose computational function has evaded inquiry, but has been implicated in a wide range of behaviours. Recent experimental evidence from the fruit fly (Drosophila melanogaster) and the cockroach (Blaberus discoidalis) has demonstrated the existence of neural activity corresponding to the animal's orientation within a virtual arena (a neural 'compass'), and this provides an insight into one component of the CX structure. There are two key features of the compass activity: an offset between the angle represented by the compass and the true angular position of visual features in the arena, and the remapping of the 270° visual arena onto an entire circle of neurons in the compass. Here we present a computational model which can reproduce this experimental evidence in detail, and predicts the computational mechanisms that underlie the data. We predict that both the offset and remapping of the fly's orientation onto the neural compass can be explained by plasticity in the synaptic weights between segments of the visual field and the neurons representing orientation. Furthermore, we predict that this learning is reliant on the existence of neural pathways that detect rotational motion across the whole visual field and uses this rotation signal to drive the rotation of activity in a neural ring attractor. Our model also reproduces the 'transitioning' between visual landmarks seen when rotationally symmetric landmarks are presented. This model can provide the basis for further investigation into the role of the central complex, which promises to be a key structure for understanding insect behaviour, as well as suggesting approaches towards creating fully autonomous robotic agents.

Humans can learn under a wide variety of feedback conditions. Reinforcement learning (RL) where a series of rewarded decisions must be made is a particularly important type of learning. Computational and behavioral studies of RL have focused mainly on Markovian decision processes where the next state depends on only the current state and action. Little is known about non Markovian decision making where the next state depends on more than the current state and action. Learning is non Markovian for example when there is no unique mapping between actions and feedback. We have produced a model based on spiking neurons that can handle these non Markovian conditions by performing policy gradient descent (Friedrich 2011). Here we examine the model’s performance and compare it with human learning and a Bayes optimal reference which provides an upper bound on performance. We find that in all cases our population of spiking neurons model well describes human performance.

In recent years Multi-Electrode Arrays (MEAs) have emerged as a powerful tool to study brain (dys) functions in-vivo and in in-vitro animal models. Typically, each session of electrophysiological experiments with such MEAs generate large amount of raw data (e.g., 60 channels/MEA, 16 bits A/D conversion, 20 kHz sampling rate: approximately 8 GB/MEA, h uncompressed) and inferring meaningful conclusions from them require rigorous and automated processing. To this goal, the current work proposes a cloud-computing based software workflow, QSpikeTools for preliminary preprocessing and analysis of neuronal activities recorded from MEAs with 60 recording sites. It exploits the facilities provided by some open-source tools to delegate CPU-intensive and independent operations to be performed on individual recorded channels (e.g., signal filtering, multi-unit activity detection, spike sorting, etc.) to a multi-core computer or a computer cluster to be executed in parallel. We report that the required time in performing the desired processing and analysis decreases significantly with increasing number of employed cores. With the commercial availability of new, sophisticated, and inexpensive high-density MEAs, we believe that widely dissemination of QSpikeTools may facilitate its adoption and customization, and possibly inspire the creation of community-supported cloud-computing facilities for MEAs users.