L'oubli est une perte temporaire ou permanente de mémoire, que l'on perçoit souvent de manière déplaisante lorsqu'elle nous empêche d'accéder à un savoir que l'on a acquis. Cependant, des découvertes récentes suggèrent que l'oubli peut aussi être un processus adaptatif permettant d'optimiser notre mémoire, en effaçant des informations non pertinentes susceptibles d'interférer avec le stockage ou le rappel de nouvelles informations. Ainsi, l'oubli adaptatif est particulièrement essentiel au fonctionnement de notre mémoire à court terme ou mémoire de travail (MT), car les informations qui y sont stockées doivent être oubliées une fois utilisées. A l'inverse, des informations peuvent être stockées pendant toute une vie dans la mémoire à long terme ou mémoire de référence (MR) chez l'animal. Les mécanismes cellulaires et moléculaires sous-tendant le stockage des informations en mémoire mais également leur oubli adaptatif restent mal connus. Au cours de cette thèse, nous avons adopté une approche comparative et utilisé trois tâches comportementales chez le rat au sein d’un même labyrinthe radial : une tâche de MR et deux tâches de MT impliquant un oubli adaptatif plus ou moins efficace. Nous avons étudié la transmission synaptique à la synapse entre le cortex entorhinal et le gyrus denté (voie d’entrée de l’hippocampe, structure clé de la mémoire) entre les deux jours d’apprentissage de ces trois tâches. Nous avons montré que la consolidation mnésique (en MR) induit un phénomène de potentialisation synaptique à long terme (proche d'une LTP), comme attendu d’après la littérature. A l'inverse, nous avons montré pour la première fois que l’oubli adaptatif en MT induirait une dépression synaptique à long terme. De plus, de nombreuses études suggèrent l’implication du sommeil dans la mémoire, mais le rôle des différentes phases de sommeil dans la consolidation mnésique ainsi que leur rôle dans l’oubli adaptatif reste ambigu. Nous avons donc également réalisé des enregistrements polysomnographiques (entre les deux jours des tâches), afin de quantifier la durée des états sommeil et la puissance des oscillations cérébrales. Nous avons ainsi confirmé le rôle du sommeil paradoxal, et plus particulièrement de ses oscillations gamma, pour la consolidation mnésique en MR. A l’inverse, l’oubli adaptatif en MT serait favorisé par les oscillations lentes du sommeil lent. Ces résultats représentent une contribution significative non seulement aux mécanismes neuronaux sous-tendant la mémoire et l’oubli adaptatif, mais également aux modulations de ces mécanismes par le sommeil. Nous avons donc montré que la consolidation mnésique induit un phénomène physiologique de potentialisation synaptique proche d'une LTP. Or l’induction artificielle de LTP par stimulation tétanique est considérée comme un modèle cellulaire de la mémoire. Notre second objectif a été d'évaluer l'impact de la modulation des états de sommeil sur une LTP, cette-fois-ci induite artificiellement (dans les mêmes conditions à la même synapse chez le rat vigile). Nous voulions ainsi comparer l'effet de la modulation du sommeil sur une potentialisation physiologique (après apprentissage) ou sur une LTP artificielle. Nous avons montré de nombreuses similitudes entre ces deux situations de potentialisation synaptique, notamment en ce qui concerne le rôle favorable du sommeil paradoxal, ce qui confirme l’intérêt de la LTP artificielle pour l’étude de la mémoire. Enfin, notre étude montre que non seulement le sommeil, mais également les oscillations de l'éveil contribuent à la mémoire et l’oubli. Nous avons analysé les oscillations locales dans le gyrus denté au cours des trois tâches comportementales déjà décrites. L'importante résolution spatiale et temporelle de cette analyse nous a permis d'identifier l'implication de certaines oscillations locales à des moments cruciaux de prise de décision dans le labyrinthe, au cours de l'encodage et du rappel d'informations stockées en MR ou en MT / Forgetting is a temporary or permanent loss of memory, often perceived as deleterious to our cognitive abilities, especially when it prevents us from accessing information we previously acquired. However, recent studies in Neurosciences suggest that forgetting could also be an adaptive phenomenon that would optimize memory function by erasing non relevant information, that could otherwise interfere with the storage or recall of new information. Therefore, adaptive forgetting would particularly be necessary for daily activities relying on short-term or working memory (WM), as information temporarily stored in WM need to be forgotten once used, so that this temporary information does not interfere in the future with the storage and recall of newer information. On the contrary, information can be stored for a lifetime in long-term or reference memory (RM) in animals. The molecular and cellular mechanisms underlying memory storage of information, but also adaptive forgetting, are still unclear. During this thesis, we used a comparative approach by training rats in three different behavioral tasks in the same radial maze: one RM task and two WM tasks involving a more or less effective adaptive forgetting process of previously stored information. We studied synaptic transmission at the synapse between the entorhinal cortex and the dentate gyrus (gating hippocampus, a key structure for memory) between two days of training in these three tasks. Our results show that memory consolidation (in RM) induces a form of long-term potentiation (LTP-like), confirming previous published work from the literature. However, we showed for the first time that adaptive forgetting in WM could trigger long-term synaptic depression. Moreover, numerous studies suggest that sleep is important for optimal memory processing, but the role of the different sleep phases in memory consolidation and in adaptive forgetting remains to be elucidated. We thus also performed polysomnographic recordings (between the two trainings days in the three behavioral tasks), in order to measure sleep state durations and sleep oscillations associated with these processes. Our results confirm the essential role of paradoxical sleep, and more specifically gamma oscillations during this state, for memory consolidation in RM. On the contrary, we also found that adaptive forgetting in WM would benefit from slow oscillations during slow wave sleep. We believe that these results contribute significantly to our understanding of neuronal mechanisms underlying memory and forgetting, especially concerning the modulation of these mechanisms by the different sleep states following training. On the one hand, we thus here showed that memory consolidation induces an LTP-like physiological phenomenon. On the other hand, the induction of an artificial form of LTP by tetanic stimulation is considered a cellular model of memory. Our second goal was to assess the modulation of an artificial LTP (at the same synapse, in the same conditions, on freely-moving rats) by sleep states. We also wanted to compare the impact of sleep states on a physiological LTP-like process (after learning) or on an artificial LTP. Our results showed many similarities between these two situations of synaptic potentiation, in particular concerning the positive role of paradoxical sleep, confirming the relevance of artificial LTP as a model to study memory processes. Finally, our study shows that not only sleep, but also oscillations during waking, could contribute to memory and forgetting. We therefore analyzed local spontaneous oscillations in the dentate gyrus while rats were performing the three behavioral tasks previously described. The high spatial and temporal resolution of this analysis allowed us to show the role of different local spontaneous oscillations at critical moments of training in the maze, in particular during decision making, and during encoding or retrieval of information stored in RM or WM

Čeněk Pýcha History in public space: Changes of institutions of memory Abstract The submitted dissertation project is based on a longer research interest in memory and remembering. Interdisciplinary memory studies is one of the most dynamically developing subdisciplines in the social sciences and humanities. The aim of this work is to contribute to the ongoing academic discussion and to explore some environments of making sense of the past, which so far stood rather on the periphery of research interests. The research field of this project is defined by the questioning of transformations of memory institutions. I observe this change primarily on the trajectory of movement from grand institutions of memory to small ones. As the grand institutions of memory, I understand the traditional institutions of the interpretation of the past that were born in the modernization process. In this dissertation project, I focus mainly on institutions of heritage preservation and museums. With the partial disintegration of grand collective frameworks, these institutions are divided into small institutions. I study this movement in case studies on contemporary cultural practices of remembrance in new memory ecologies. I focus on digital platforms for travelers, remembering through visual communication or interest in places...

While deep learning has proved to be useful in a large variety of tasks, a limitation remains of needing all classes and samples to be present at the training stage in supervised problems. This is a major issue in the field of biomedical imaging since keeping samples in the training sets consistently is often a liability. Furthermore, this issue prevents the simple updating of older models with only the new data when it is introduced, and prevents collaboration between companies. In this work, we examine an array of Continual Learning approaches to try to improve upon the baseline of the naive finetuning approach when retraining on new tasks, and achieve accuracy levels similar to the ones seen when all the data is available at the same time. Continual learning approaches with which we attempt to mitigate the problem are EWC, UCB, EWC Online, SI, MAS, CN-DPM. We explore some complex scenarios with varied classes being included in the tasks, as well as close to ideal scenarios where the sample size is balanced among the tasks. Overall, we focus on X-ray images, since they encompass a large variety of diseases, with new diseases requiring retraining. In the preferred setting, where classes are relatively balanced, we get an accuracy of 63.30 versus a baseline of 53.92 and the target score of 66.83. For the continued training on the same classes, we get an accuracy of 35.52 versus a baseline of 27.73. We also examine whether learning rate adjustments at task level improve accuracy, with some improvements for EWC Online. The preliminary results indicate that CL approaches such as EWC Online and SI could be integrated into radiography data learning pipelines to reduce catastrophic forgetting in situations where some level of sequential training ability justifies the significant computational overhead. / Även om djupinlärning har visat sig vara användbart i en mängd olika uppgifter, kvarstår en begränsning av att behöva alla klasser och prover som finns på utbildningsstadiet i övervakade problem. Detta är en viktig fråga inom området biomedicinsk avbildning eftersom det ofta är en belastning att hålla prover i träningsuppsättningarna. Dessutom förhindrar det här problemet enkel uppdatering av äldre modeller med endast nya data när de introduceras och förhindrar samarbete mellan företag. I det här arbetet undersöker vi en rad kontinuerliga inlärningsmetoder för att försöka förbättra baslinjen för den naiva finjusteringsmetoden vid omskolning på nya uppgifter och närma sig noggrannhetsnivåer som de som ses när alla data är tillgängliga samtidigt. Kontinuerliga inlärningsmetoder som vi försöker mildra problemet med inkluderar bland annat EWC, UCB, EWC Online, SI. Vi utforskar några komplexa scenarier med olika klasser som ingår i uppgifterna, samt nära idealiska scenarier där exempelstorleken balanseras mellan uppgifterna. Sammantaget fokuserar vi på röntgenbilder, eftersom de omfattar ett stort antal sjukdomar, med nya sjukdomar som kräver omskolning. I den föredragna inställningen får vi en noggrannhet på 63,30 jämfört med en baslinje på 53,92 och målpoängen på 66,83. Medan vi för den utökade träningen på samma klasser får en noggrannhet på 35,52 jämfört med en baslinje på 27,73. Vi undersöker också om justeringar av inlärningsfrekvensen på uppgiftsnivå förbättrar noggrannheten, med vissa förbättringar för EWC Online. De preliminära resultaten tyder på att CL-metoder som EWC Online och SI kan integreras i rörledningar för röntgendatainlärning för att minska katastrofal glömska i situationer där en viss nivå av sekventiell utbildningsförmåga motiverar den betydande beräkningskostnaden.

Medical doctors understaffing is becoming a compelling problem in many healthcare systems. This problem can be alleviated by utilising Computer-Aided Diagnosis (CAD) systems to substitute doctors in different tasks, for instance, histopa-thological image classification. The recent surge of deep learning has allowed CAD systems to perform this task at a very competitive performance. However, a major challenge with this task is the need to periodically update the models with new data and/or new classes or diseases. These periodical updates will result in catastrophic forgetting, as Convolutional Neural Networks typically requires the entire data set beforehand and tend to lose knowledge about old data when trained on new data. Incremental learning methods were proposed to alleviate this problem with deep learning. In this thesis, two incremental learning methods, Learning without Forgetting (LwF) and a generative rehearsal-based method, are investigated. They are evaluated on two criteria: The first, capability of incrementally adding new classes to a pre-trained model, and the second is the ability to update the current model with an new unbalanced data set. Experiments shows that LwF does not retain knowledge properly for the two cases. Further experiments are needed to draw any definite conclusions, for instance using another training approach for the classes and try different combinations of losses. On the other hand, the generative rehearsal-based method tends to work for one class, showing a good potential to work if better quality images were generated. Additional experiments are also required in order to investigating new architectures and approaches for a more stable training.

The master thesis focuses on similarities and disparities in the politics of memory related to the 'forgotten' Romani Holocaust in Slovakia and Hungary after 1989. It scrutinizes based on particular topic areas to what extent is the Romani Holocaust marginalized, excluded or integrated into the historical narratives of nation-states and whether it can be classified as a competitive or multidirectional form of memory. The reference point for the comparative discursive analysis represents the Jewish Holocaust. By employing the discourse-historical approach, the thesis attempts to identify various current or long-term strategies of instrumentalizing the Romani Holocaust in specific politics of memory. Moreover, it endeavours to find out which role can the Romani Holocaust play in forming the national identity in these states or its potential endangering.

Dans l’apprentissage fédéré, un modèle global est appris en agrégeant les mises à jour du modèle calculées à partir d’un ensemble de nœuds clients, un défi clé dans ce domaine est l’hétérogénéité des données entre les clients qui dégrade les performances du modèle. Les algorithmes d’apprentissage fédéré standard effectuent plusieurs étapes de gradient avant de synchroniser le modèle, ce qui peut amener les clients à minimiser exagérément leur propre objectif local et à s’écarter de la solution globale. Nous démontrons que dans un tel contexte, les modèles de clients individuels subissent un oubli catastrophique par rapport aux données d’autres clients et nous proposons une approche simple mais efficace qui modifie l’objectif d’entropie croisée sur une base par client en repondérant le softmax de les logits avant de calculer la perte. Cette approche protège les classes en dehors de l’ensemble d’étiquettes d’un client d’un changement de représentation brutal. Grâce à une évaluation empirique approfondie, nous démontrons que notre approche peut atténuer ce problème, en apportant une amélioration continue aux algorithmes d’apprentissage fédéré standard. Cette approche est particulièrement avantageux dans les contextes d’apprentissage fédéré difficiles les plus étroitement alignés sur les scénarios du monde réel où l’hétérogénéité des données est élevée et la participation des clients à chaque cycle est faible. Nous étudions également les effets de l’utilisation de la normalisation par lots et de la normalisation de groupe avec notre méthode et constatons que la normalisation par lots, qui était auparavant considérée comme préjudiciable à l’apprentissage fédéré, fonctionne exceptionnellement bien avec notre softmax repondéré, remettant en question certaines hypothèses antérieures sur la normalisation dans un système fédéré / In Federated Learning, a global model is learned by aggregating model updates computed from a set of client nodes, a key challenge in this domain is data heterogeneity across clients which degrades model performance. Standard federated learning algorithms perform multiple gradient steps before synchronizing the model which can lead to clients overly minimizing their own local objective and diverging from the global solution. We demonstrate that in such a setting, individual client models experience a catastrophic forgetting with respect to data from other clients and we propose a simple yet efficient approach that modifies the cross-entropy objective on a per-client basis by re-weighting the softmax of the logits prior to computing the loss. This approach shields classes outside a client’s label set from abrupt representation change. Through extensive empirical evaluation, we demonstrate our approach can alleviate this problem, providing consistent improvement to standard federated learning algorithms. It is particularly beneficial under the challenging federated learning settings most closely aligned with real world scenarios where data heterogeneity is high and client participation in each round is low. We also investigate the effects of using batch normalization and group normalization with our method and find that batch normalization which has previously been considered detrimental to federated learning performs particularly well with our re-weighted softmax, calling into question some prior assumptions about normalization in a federated setting

No description available.

No description available.

No description available.

No description available.