In the course of the twentieth century, the representation of the relationship between time, narrative, and memory in French literature undergoes a change. Proust's <i>A la recherche du temps perdu</i> evidences the power of memory to synthesize separate moments through a work of art and embodies the thesis of Paul Ricoeur on the interdependency of time and narrative. This proustian synthesis becomes almost impossible for Samuel Beckett, George Perec, and Maurice Blanchot, who encounter the disappearance of memory preserved in the narrative space. Not only in the face of the spatialization of memory in the increasing development of mass media, which Walter Benjamin criticizes, but also in the face of the deterioration of space, the translation of a temporalized memory into a coherent story cannot continue. From Proust's immense edifice of memory, to Beckett's invention, to Perec's spatial preservation of vanished time, to Blanchot's désoeuvrement, the works of these authors traverse the wide span between memory and forgetting in literature. This paper integrates these authors who diverge in the direction of memory, on the one hand, and in the direction of forgetting, on the other. This integration forms passageways that make possible the exchange between these two ends. Communication continues in the gap where memory and forgetting are reflections of each other. / Au cours du vingtième siècle, la représentation du rapport entre le temps, le récit, et la mémoire dans la littérature française subit un changement. A la recherche du temps perdu de Proust atteste d’une mémoire puissante qui synthétise des moments séparés dans une oeuvre d’art, et matérialise la thèse de Paul Ricoeur au sujet de l’interdépendance du temps et du récit. Cette synthèse proustienne devient presque impossible chez Samuel Beckett, Georges Perec, et Maurice Blanchot, qui affrontent la disparition de la mémoire préservée dans l’espace narrative. La traduction de la mémoire temporalisée dans une histoire cohérente ne peut pas continuer non seulement face à la spatialisation de la mémoire dans l’accroissement des développements des médias de masse que Walter Benjamin critique, mais aussi face à la détérioration de l’espace. De l’immense édifice de la mémoire chez Proust, à l’invention de Beckett, à la préservation spatiale du temps disparu chez Perec, et au désoeuvrement de Blanchot, les oeuvres de ces auteurs franchissent la vaste étendue entre la mémoire et l’oubli dans la littérature. La présente analyse intègre des auteurs qui se divergent au côté de la mémoire et au côté de l’oubli. Cette intégration établit des passages qui rendent possible l’échange entre ces deux côtés. La communication continue dans l’écart où la mémoire et l’oubli se reflètent. / Master of Arts

Generalized collaborative drones are a technology that has many potential benefits. General purpose drones that can handle exploration, navigation, manipulation, and more without having to be reprogrammed would be an immense breakthrough for usability and adoption of the technology. The ability to develop these multi-task, multi-agent drone systems is limited by the lack of available training environments, as well as deficiencies of multi-task learning due to a phenomenon known as catastrophic forgetting. In this thesis, we present a set of simulation environments for exploring the abilities of multi-task drone systems and provide a platform for testing agents in incremental single-agent and multi-agent learning scenarios. The multi-task platform is an extension of an existing drone simulation environment written in Python using the PyBullet Physics Simulation Engine, with these environments incorporated. Using this platform, we present an analysis of Incremental Learning and detail the beneficial impacts of using the technique for multi-task learning, with respect to multi-task learning speed and catastrophic forgetting. Finally, we introduce a novel algorithm, Incremental Learning with Second-Order Approximation Regularization (IL-SOAR), to mitigate some of the effects of catastrophic forgetting in multi-task learning. We show the impact of this method and contrast the performance relative to a multi-agent multi-task approach using a centralized policy sharing algorithm. / Master of Science / Machine Learning techniques allow drones to be trained to achieve tasks which are otherwise time-consuming or difficult. The goal of this thesis is to facilitate the work of creating these complex drone machine learning systems by exploring Reinforcement Learning (RL), a field of machine learning which involves learning the correct actions to take through experience. Currently, RL methods are effective in the design of drones which are able to solve one particular task. The next step in this technology is to develop RL systems which are able to handle generalization and perform well across multiple tasks. In this thesis, simulation environments for drones to learn complex tasks are created, and algorithms which are able to train drones in multiple hard tasks are developed and tested. We explore the benefits of using a specific multi-task training technique known as Incremental Learning. Additionally, we consider one of the prohibitive factors of multi-task machine learning-based solutions, the degradation problem of agent performance on previously learned tasks, known as catastrophic forgetting. We create an algorithm that aims to prevent the impact of forgetting when training drones sequentially on new tasks. We contrast this approach with a multi-agent solution, where multiple drones learn simultaneously across the tasks.

Language modeling, especially through the use of transformer-based large language models (LLMs), has drastically changed how we view and use artificial intelligence (AI) and machine learning (ML) in our daily lives. Although LLMs have showcased remarkable linguistic proficiency in their abilities to write, summarize, and phrase, these model have yet to achieve the same remarkability in their ability to quantitatively reason. This deficiency is specially apparent in smaller models (less than 1 Billion parameters) than can run natively on-device. Between the complementary capabilities of qualitative and quantitative reasoning, this thesis focuses on the latter, where the goal is to devise mechanisms to instill quantitative reasoning capabilities into these models. However, instilling this notion is not as straight forward as traditional end-to-end learning. The learning of quantitative notions include the ability of the model to discern between regular linguistic tokens and magnitude/scale-oriented non-linguistic tokens. The learning of these notions, specially after pre-training, comes at a cost for these models: catastrophic forgetting. Thus, learning needs to be followed with retention - making sure these models do not forget what they have learned. Thus, we first motivate the need for numeracy-enhanced models via their potential applications in field of data-to-text generation (D2T), showcasing how these models behave as quantitative reasoners as-is. Then, we devise both token-level training interventions and information-theoretic training interventions to numerically enhance these models, with the latter specifically focused on combating catastrophic forgetting. Our information-theoretic interventions not only lead to numerically-enhanced models but lend us critical insights into the learning behavior of these models, especially when it comes to adapting these models to the target task distribution from their pretraining distribution. Finally, we extrapolate these insights to devise more effective strategies transfer learning and unlearning for language modeling. / Doctor of Philosophy / Language modeling, especially through the use of transformer-based large language models (LLMs), has drastically changed how we view and use artificial intelligence (AI) and machine learning (ML) in our daily lives. Although LLMs have showcased remarkable linguistic proficiency in their abilities to write, summarize, and phrase, these model have yet to achieve the same remarkability in their ability to quantitatively reason. This deficiency is specially apparent in smaller models than can run natively on-device. This thesis focuses on instilling within these models the ability to perform quantitative reasoning - the ability to differentiate between words and numbers and understand the notions of magnitude tied with said numbers, while retaining their linguistic skills. The learned insights from our experiments are further used to devise models that better adapt to target tasks.

We all experience moments in our life, that we would rather forget. Some people are even haunted by unpleasant memories, that involuntarily intrude into their awareness and harm their well-being. Intrusions - unwanted, unintended, recurrent memories - are a hallmark symptom in psychological disorders, like post-traumatic stress disorder, depression, or anxiety disorder. Fortunately, we are not fully at the mercy of involuntary retrieval processes. Research over the last two decades has shown that active suppression of memory retrieval can eventually lead to forgetting. Memories that have repeatedly been suppressed, are more often forgotten later on than control memories. This phenomenon is called suppression-induced forgetting. Neuroimaging studies have revealed that during suppression the dorsolateral prefrontal cortex shuts down retrieval processes in the hippocampus and other memory-related regions. However, the sustained effects of suppression on the memory remain unknown. The current thesis aims to close this gap and characterize the behavioral, neural, and emotional aftereffects of memory suppression. (i) A large meta-analysis confirms that suppression-induced forgetting is a robust phenomenon in healthy participants. Moreover, it indicates that this mechanism is deficient in participants with (sub)clinical manifestations of depression and anxiety. This suggests that memory suppression is an adaptive mechanism, that helps maintain our well- being. (ii) A replication and mini-meta-analysis corroborate that suppression can even weaken complex aversive scene memories. The employed methodological procedure allows us to study the effects of suppression in a fine-grained and naturalistic way. (iii) We thus adapted the procedure in an fMRI study and investigated for the first time the neural aftereffects of suppression. We show that suppression renders memories less vivid and that this reduction in vividness is associated with reduced reinstatement of the individual neural memory trace in the parahippocampal cortex. After suppression, we find overall less evidence for the processing of complex aversive scenes across the brain. These findings indicate that suppression has a sustained effect on neural memory traces and subsequently leads to impoverished reinstatement of the neural memory trace. (iv) Finally, we investigated whether suppression can weaken emotional responses that are associated with unwanted memories. In a preregistered psychophysiological study, we simultaneously tested the effect of suppression on episodic fear memories, as well as associated conditioned fear responses. We show that fear memories become less accessible after suppression, but did not find the predicted weakening of the emotional response. Future studies will have to follow up on this question. The present thesis demonstrates that memory suppression has a sustained weakening effect on the behavioral and neural levels. Our findings also indicate that suppression is a powerful and adaptive mechanism that supports our well-being. The results are of particular relevance as deficient suppression may play a critical role in both the development and maintenance of psychological disorders.

本研究主要目的在於運用指示遺忘作業探討高社交焦慮者是否有記憶偏誤現象。高社交焦慮組24人與低社交焦慮組20人參與正式實驗，受試者被要求依據指示記住或忘記三種類型刺激詞（社交威脅、中性、社交正向）。作業結果未發現任何顯著之組間差異，但進一步分析發現，在指示記住項目的回憶上，所有受試者皆回憶出較多的社交正向詞（相對於中性詞以及社交威脅詞），顯示記憶的正向偏誤，然而，在指示忘記項目的回憶上，正向偏誤的現象僅出現在低社交焦慮組，不見於高社交焦慮組。此外，相關分析顯示，高社交焦慮組的FNE量表（Fear of Negative Evaluation Scale）得分與指示記住的社交正向詞回憶量成正相關，與指示忘記的社交威脅詞回憶量成負相關；低社交焦慮組的FNE量表得分與指示記住的社交正向詞回憶量成負相關。這些結果顯示高社交焦慮者可能缺乏正向偏誤的保護機制，較一般人更容易遺忘正向評價訊息，因而更容易受到負面評價的影響，於是他們傾向抑制負向評價訊息的回憶，將注意力投注在非威脅訊息，並努力記住正向評價，換言之，他們透過逃避負面評價的方式來維持良好的自我形象。 / The purpose of the present study was to utilize the directed forgetting task to investigate the memory bias in socially anxious individuals. Performance on a directed forgetting task was assessed in socially anxious (n=24) and nonanxious (n=20) individuals. Participants were presented with three types of words (negative social, neutral, positive social) and were cued to either remember or forget each word as it was presented. There were no between-groups differences on a free recall task for words in both remember and forget conditions. Follow-up analyses demonstrated that all subjects recalled more positive social words than either neutral words or negative social words in the remember condition, revealing the positive memory bias. However, only nonanxious individuals showed such positive memory bias in the forget condition, while high-social-anxiety individuals did not. Moreover, in the social anxiety group, the Fear of Negative Evaluation Scale (FNE) scores were positively related to the recall of positive social “to-be-remembered” words, and negatively related to the recall of negative social “to-be-forgotten” words. In the control group, FNE scores were negatively related to the recall of positive social “to-be-remembered” words. These results suggest that the protective positive bias was absent in socially anxious individuals. Thus, they were more likely than nonanxious individuals to forget positive evaluative information; therefore, they were vulnerable to negative evaluation. Furthermore, people with excessive social anxiety might try hard to inhibit the recall of negative evaluation and to remember positive evaluation. In other words, they made an effort to maintain a good impression by avoiding negative evaluative information.

Esta pesquisa trata do esquecimento do passado por refugiados africanos. As entrevistas foram recolhidas na Casa do Migrante, albergue que acolhe migrantes internos, imigrantes e refugiados recém-chegados em São Paulo. Foram entrevistados dois africanos: um angolano e outro congolês. Nosso objetivo de estudar o esquecimento emergiu nas entrevistas individuais e compartilhada entre estes refugiados, pois eles não queriam lembrar as cenas de guerra em África. Compreendemos o esquecimento, levantando a hipótese freudiana de que os refugiados querem esquecer o passado pois, ao tentarem dominar o golpe excessivo, repetem compulsivamente o trauma e a hipótese benjaminiana de que a dificuldade dos africanos em comunicar a experiência de guerra se deve ao declínio da narrativa e a experiência do choque após o avanço das forças produtivas. Porém, buscamos investigar se é possível elaborar o passado. Compreendemos as levas de refugiados ao redor do mundo como resultado da crise do capitalismo global, como nos mostrou Robert Kurz. Não se trata de povos obrigados a sair de sua pátria desde a antiga história religiosa da humanidade, tão pouco de vítimas de perseguição ou vítimas de violação dos direitos humanos, como concebe a Cáritas Arquidiocesana de São Paulo no atendimento aos refugiados. Analisamos as entrevistas a partir de três categorias de análise a fuga da guerra, a educação e o trabalho. Então, refletimos sobre o esquecimento dos refugiados africanos, partindo de um teor religioso para alcançar algumas considerações psicológicas. Mas o que resta aos psicanalistas diante de refugiados africanos? Concluímos o estudo, investigando a metodologia psicanalítica mais adequada para a pesquisa com africanos sobreviventes de guerra. E decidimos recolher seus sonhos traumáticos, segundo nossa hipótese de que eles pudessem sonhar à noite com aquilo que querem esquecer à luz do dia / This study is about the forgetting of the past by African refugees. The interviews were collected at Casa do Migrante hostel that offers shelter for migrants, immigrants and newcomers refugees in São Paulo. Two Africans were interviewed: an Angolan and a Congolese. Our aim of studying the forgetting emerged from the individual and shared interviews with these refugees, because they did not want to remember the scenes of war in Africa. We understand the forgetting, considering the freudian hypothesis that the refugees want to forget the past, as they compulsively repeat the trauma, when they try to dominate the excessive coup and considering the benjaminian hypothesis that the difficulty of Africans to communicate the war experience is due to the decline of narrative and the shock experience after the development of productive forces. Nevertheless, we try to investigate whether it is possible to work through the past. We understand the waves of refugees around the world as a result of the crisis of global capitalism, as Robert Kurz showed us. It is not about people obliged to leave home since the ancient religious history of mankind, or about victims of persecution or victims of human rights violation, as conceived by the Cáritas Arquidiocesana de São Paulo in the attendance of refugees. We analyze the interviews according to three categories of analysis the flight from war, education and work. So, we thought about the forgetting of African refugees, starting from a religious content to achieve some psychological considerations. But what does it remain for psychoanalysts in the presence of African refugees? We conclude the study, investigating the more suitable psychoanalytic methodology for the research with Africans war survivors. And we decide to collect their traumatic dreams, according to our hypothesis that they might dream at night with what they want to forget at day light

The present experiment was conducted to investigate the effects of reinforcer magnitude on resistance to disruption of remembering and response rates. Pigeons were exposed to a variable-interval (VI), delayed-matching-to-sample procedure (DMTS) with two components (rich and lean). Specifically, completion of a VI 20 second (s) multiple schedule resulted in DMTS trials in both components. In a DMTS trial, a choice of one of two comparison stimuli (e.g., blue key) results in reinforcement if the choice matches some property of the sample stimulus presented previously. Sample and comparison stimuli are separated by a delay. Four delays (0.1, 4, 8, and 16 s) were used between the sample and comparison stimuli in the study. The difference between rich and lean components was the length of hopper duration following a correct response. The probability of reinforcement following a correct response in both components was .5. Each pigeon was exposed to 50 sessions of initial baseline and then 30 sessions of baseline between each disruptive condition (extinction, intercomponent interval [ICI] food, lighting the houselight during delays, and prefeeding). Separable aspects of the forgetting functions (initial discriminability and rate of forgetting) were examined by determining accuracy at each delay. During baseline, response rates were higher in the rich component relative to the lean. Accuracy decreased as delay increased in both rich and lean components, and accuracy was consistently higher in the rich relative to the lean component. During disruptive conditions, extinction, ICI food, and prefeeding disrupted response rates, but lighting the houselight during the delays had little effect. During the DMTS portion of the procedure, extinction and prefeeding decreased initial discriminability and lighting the houselight during the delay increased rate of forgetting. Intercomponent food had little effect on accuracy. Accuracy in the rich component was more resistant to disruption relative to the lean component during extinction. These results indicate that certain disruptors do not have the same disruptive effect across response rates and accuracy (e.g., ICI food). These data also suggest that when systematic differences in accuracy between rich and lean components are revealed, performance in the rich component tends to be more resistant to disruption.

<p>This thesis presents four studies investigating (1) whether incidentally primed control-related words can attenuate the impact of activated stereotypes on subsequent evaluation of a target person, (2) the impact of motivated forgetting on the recall of stereotypically congruent and incongruent information, and (3) the impact of a directed forgetting instruction on the false recall and recognition of nonpresented stereotypical information.</p><p>In three experiments, Study I showed that participants initially primed with the social category, <i>immigrant, </i>and subsequently primed with words that were evocative of control or self-control made less negative impression of a target displaying ambiguous behaviors than participants not exposed to such words.</p><p>Study II, using a directed-forgetting paradigm, demonstrated in two experiments that participants subliminally primed with Swedish facial photographs who later studied stereotypically incongruent words roughly recalled an equal number of items regardless of the forget or remember instructions. </p><p>Study III showed that participants primed with the social category, <i>immigrant</i> and then studied a list of stereotypically related and unrelated words falsely recognized more nonpresented stereotypical words when they were furnished with a forget than a remember instruction. Similarly, Study IV (Experiment 2) demonstrated that participants primed with the social category, <i>immigrant</i>, but not with a neutral category, falsely recalled more nonpresented stereotypical words when their cognitive capacity was depleted through a concurrent memory load task. </p><p>The thesis presents a review and a discussion of some of the theoretical underpinnings of the extant literature on stereotyping and intergroup relations and of the social implications of the present findings.</p>

This thesis presents four studies investigating (1) whether incidentally primed control-related words can attenuate the impact of activated stereotypes on subsequent evaluation of a target person, (2) the impact of motivated forgetting on the recall of stereotypically congruent and incongruent information, and (3) the impact of a directed forgetting instruction on the false recall and recognition of nonpresented stereotypical information. In three experiments, Study I showed that participants initially primed with the social category, immigrant, and subsequently primed with words that were evocative of control or self-control made less negative impression of a target displaying ambiguous behaviors than participants not exposed to such words. Study II, using a directed-forgetting paradigm, demonstrated in two experiments that participants subliminally primed with Swedish facial photographs who later studied stereotypically incongruent words roughly recalled an equal number of items regardless of the forget or remember instructions. Study III showed that participants primed with the social category, immigrant and then studied a list of stereotypically related and unrelated words falsely recognized more nonpresented stereotypical words when they were furnished with a forget than a remember instruction. Similarly, Study IV (Experiment 2) demonstrated that participants primed with the social category, immigrant, but not with a neutral category, falsely recalled more nonpresented stereotypical words when their cognitive capacity was depleted through a concurrent memory load task. The thesis presents a review and a discussion of some of the theoretical underpinnings of the extant literature on stereotyping and intergroup relations and of the social implications of the present findings.

The current study investigated the effects of aging on the strategic control of attention at encoding and the extent to which this relationship was mediated by working memory capacity. The value-directed remembering task used by Castel et al. (2009) was modified to include an inhibitory task demand (i.e., value-directed forgetting), and age differences were predicted due to declines in the efficiency of inhibitory mechanisms. Results confirmed this prediction, as older adults were less efficient in maximizing their selectivity scores upon the inclusion of task interference, and working memory was found to be supportive of performance. Results additionally support an age-related decline in the directed forgetting effect, such that older adults recalled and recognized fewer TBR items and more TBF items, relative to younger adults. Taken together, results suggest an age-related decline in the ability to inhibit goal-irrelevant information, thereby limiting working memory resources available for greater processing of goal-relevant information.