This work includes an attempt to enhance the autonomy of intelligent agents via real-time learning.In nature, the ability to learn at runtime gives species which can do so key advantages over others. While most AI systems do not need to have this ability but can be trained before deployment, it allows agents to adapt, at runtime, to changing and generally unknown circumstances, and then to exploit their environment for their own purposes. To reach this goal, in this thesis a pragmatic design (ONA) for a general-purpose reasoner incorporating Non-Axiomatic Reasoning System (NARS) theory is explored. The design and implementation is presented in detail, in addition to the theoretical foundation. Then, experiments related to various system capabilities are carried out and summarized, together with application projects where ONA is utilized: a traffic surveillance application in the Smart City domain to identify traffic anomalies through real-time reasoning and learning, and a system to help first responders by providing driving assistance and presenting of mission-critical information. Also it is shown how reliable real-time learning can help to increase autonomy of intelligent agents beyond the current state-of-the-art. Here, theoretical and practical comparisons with established frameworks and specific techniques such as Q-Learning are made, and it is shown that ONA does also work in non-Markovian environments where Q-Learning cannot be applied. Some of the reasoner's capabilities are also demonstrated on real robotic hardware. The experiments there show combining learning knowledge at runtime with the utilization of only partly complete mission-related background knowledge given by the designer, allowing the agent to perform a complex task from an only minimal mission specification which does not include learnable details. Overall, ONA is suitable for autonomous agents as it combines, in a single technique, the strengths of behavior learning, which is usually captured by Reinforcement Learning, and means-end reasoning (such as Belief-Desire-Intention models with planner) to effectively utilize knowledge expressed by a designer. / Computer and Information Science

Estudos sobre o papel do responder por exclusão na aquisição de vocabulário têm se concentrado nas relações nome-objeto ou nome-figura, mas não está claro se esse processo está envolvido, e de que maneira na aprendizagem de diferentes categorias lexicais, adjetivos (propriedades do objeto), por exemplo. Este estudo teve por objetivo comparar a aprendizagem de relações nome-textura por participantes com diferentes perfis de desenvolvimento (crianças com desenvolvimento típico e indivíduos com deficiência intelectual) em dois procedimentos de ensino por exclusão. Participaram deste estudo 24 indivíduos, 12 com desenvolvimento típico (Grupo DT) e idade entre 5 e 6 anos e, 12 com deficiência intelectual com idade entre 8 a 13 anos (Grupo DI), todos com nível de desenvolvimento linguístico entre 5 e 6 anos. Os estímulos utilizados foram palavras ditadas (nomes de diferentes texturas) e estímulos táteis (objetos confeccionados com diferentes texturas, referentes aos nomes ditados). O procedimento tinha quatro fases. A primeira investigou se os participantes conheciam os nomes de texturas familiares (áspero e liso) linha de base. Na segunda fase, foram ensinadas quatro palavras desconhecidas (referentes a nomes de texturas), em duas condições. Na primeira, duas texturas (rugoso e aveludado) eram ensinadas com o uso de contraste linguístico referencial (CAC). Na segunda condição, outras duas texturas (listado e flocado) foram ensinadas em um procedimento de ensino por exclusão padrão (SAC). O critério de aprendizagem era 100% de acertos no bloco de sondas; blocos de ensino eram conduzidos até se atingir esse critério. Uma semana após cada condição, eram conduzidas sondas de manutenção dos nomes de adjetivos ensinados. Na terceira e quarta fase, sondas de manutenção geral (SMG) e sondas de generalização - SG (objetos diferentes com as mesmas texturas ensinadas) eram realizadas, bem como sondas de nomeação, apresentando-se os seis objetos com as diferentes texturas ensinadas. Ambos os grupos apresentaram desempenho similar: na Condição CAC, os participantes precisaram, em média, de um bloco de ensino para atingir o critério de aprendizagem; na Condição SAC precisaram de, em média, dois blocos. A Condição CAC favoreceu a aprendizagem das palavras novas, a partir de um número menor de blocos de ensino, porém, a exposição mínima, nas duas condições dificultou a manutenção e generalização dessa aprendizagem. Discute-se o papel do contraste linguístico referencial como dica contextual na aprendizagem de adjetivos em um procedimento de ensino por exclusão. / Studies about the role of responding by exclusion in vocabulary acquisition have focused on relations object name or name-figure, but it is unclear whether this process is involved, and how the learning of different lexical categories, adjectives (properties object), for example. This study aimed to compare the learning of name-texture relations by participants with different development profiles (children with typical development and individuals with intellectual disability) in two teaching procedures for exclusion. The study included 24 subjects, 12 typically developing (DT Group) and aged between 5 and 6 years, and 12 with intellectual disabilities 8-13 years aged (DI Group), all with level of language development between 5 and 6 years. The stimuli used were dictated words (names of different textures) and tactile stimuli (objects made with different textures, referring to the dictates names). The procedure had four phases. The first investigated whether the participants knew the names of familiar textures (rough and smooth) - baseline. In the first two textures (rough and smooth) were taught using referential linguistic contrast (CAC). In the second condition, another two textures (wrinkled and velvety) were taught in procedure teaching by exclusion (SAC). The criterion for learning was 100% correct in probes block; teaching blocks were conducted to achieve this criterion. One week after each condition, maintenance probes were conducted of the names of adjectives taught. In the third and fourth phase, general maintenance probes (SMG) and generalization probes - SG (different objects with the same taught textures) were held, as well as appointment of probes, presenting the six objects with different textures taught. Both groups had similar performance: the CAC condition, participants had to, on average, a teaching block to reach the criterion of learning; SAC on condition needed on average two blocks. The Condition CAC favored the learning of new words from a smaller number of teaching blocks, however, the minimal exposure, in both conditions difficult to maintenance and generalization of this learning. It discusses the role of the reference linguistic contrast to contextual clue in learning adjectives in a teaching procedure for exclusion.

Estudos sobre o papel do responder por exclusão na aquisição de vocabulário têm se concentrado nas relações nome-objeto ou nome-figura, mas não está claro se esse processo está envolvido, e de que maneira na aprendizagem de diferentes categorias lexicais, adjetivos (propriedades do objeto), por exemplo. Este estudo teve por objetivo comparar a aprendizagem de relações nome-textura por participantes com diferentes perfis de desenvolvimento (crianças com desenvolvimento típico e indivíduos com deficiência intelectual) em dois procedimentos de ensino por exclusão. Participaram deste estudo 24 indivíduos, 12 com desenvolvimento típico (Grupo DT) e idade entre 5 e 6 anos e, 12 com deficiência intelectual com idade entre 8 a 13 anos (Grupo DI), todos com nível de desenvolvimento linguístico entre 5 e 6 anos. Os estímulos utilizados foram palavras ditadas (nomes de diferentes texturas) e estímulos táteis (objetos confeccionados com diferentes texturas, referentes aos nomes ditados). O procedimento tinha quatro fases. A primeira investigou se os participantes conheciam os nomes de texturas familiares (áspero e liso) linha de base. Na segunda fase, foram ensinadas quatro palavras desconhecidas (referentes a nomes de texturas), em duas condições. Na primeira, duas texturas (rugoso e aveludado) eram ensinadas com o uso de contraste linguístico referencial (CAC). Na segunda condição, outras duas texturas (listado e flocado) foram ensinadas em um procedimento de ensino por exclusão padrão (SAC). O critério de aprendizagem era 100% de acertos no bloco de sondas; blocos de ensino eram conduzidos até se atingir esse critério. Uma semana após cada condição, eram conduzidas sondas de manutenção dos nomes de adjetivos ensinados. Na terceira e quarta fase, sondas de manutenção geral (SMG) e sondas de generalização - SG (objetos diferentes com as mesmas texturas ensinadas) eram realizadas, bem como sondas de nomeação, apresentando-se os seis objetos com as diferentes texturas ensinadas. Ambos os grupos apresentaram desempenho similar: na Condição CAC, os participantes precisaram, em média, de um bloco de ensino para atingir o critério de aprendizagem; na Condição SAC precisaram de, em média, dois blocos. A Condição CAC favoreceu a aprendizagem das palavras novas, a partir de um número menor de blocos de ensino, porém, a exposição mínima, nas duas condições dificultou a manutenção e generalização dessa aprendizagem. Discute-se o papel do contraste linguístico referencial como dica contextual na aprendizagem de adjetivos em um procedimento de ensino por exclusão. / Studies about the role of responding by exclusion in vocabulary acquisition have focused on relations object name or name-figure, but it is unclear whether this process is involved, and how the learning of different lexical categories, adjectives (properties object), for example. This study aimed to compare the learning of name-texture relations by participants with different development profiles (children with typical development and individuals with intellectual disability) in two teaching procedures for exclusion. The study included 24 subjects, 12 typically developing (DT Group) and aged between 5 and 6 years, and 12 with intellectual disabilities 8-13 years aged (DI Group), all with level of language development between 5 and 6 years. The stimuli used were dictated words (names of different textures) and tactile stimuli (objects made with different textures, referring to the dictates names). The procedure had four phases. The first investigated whether the participants knew the names of familiar textures (rough and smooth) - baseline. In the first two textures (rough and smooth) were taught using referential linguistic contrast (CAC). In the second condition, another two textures (wrinkled and velvety) were taught in procedure teaching by exclusion (SAC). The criterion for learning was 100% correct in probes block; teaching blocks were conducted to achieve this criterion. One week after each condition, maintenance probes were conducted of the names of adjectives taught. In the third and fourth phase, general maintenance probes (SMG) and generalization probes - SG (different objects with the same taught textures) were held, as well as appointment of probes, presenting the six objects with different textures taught. Both groups had similar performance: the CAC condition, participants had to, on average, a teaching block to reach the criterion of learning; SAC on condition needed on average two blocks. The Condition CAC favored the learning of new words from a smaller number of teaching blocks, however, the minimal exposure, in both conditions difficult to maintenance and generalization of this learning. It discusses the role of the reference linguistic contrast to contextual clue in learning adjectives in a teaching procedure for exclusion.

<p dir="ltr">Today’s deep neural networks (DNNs) have achieved tremendous performance in various domains such as computer vision, natural language processing, robotics, generative tasks etc. However, these high-performing DNNs require enormous amounts of compute, resulting in significant power consumption. Moreover, these often struggle in terms of visual understanding capabilities. To that effect, this thesis focuses on two aspects - enhancing efficiency of neural networks and improving their visual understanding. On the efficiency front, we leverage brain-inspired Spiking Neural Networks (SNNs), which offer a promising alternative to traditional deep learning. We first perform a comparative analysis between models with and without leak, revealing that leaky-integrate-and-fire (LIF) model provides improved robustness and better generalization compared to integrate-and-fire (IF). However, leak decreases the sparsity of computation. In the second work, by introducing a Discrete Cosine Transform-based novel spike encoding scheme (DCT-SNN), we demonstrate significant performance improvements, achieving 2-14X reduction in latency compared to state-of-the-art SNNs. Next, a novel temporal pruning method is proposed, which dynamically reduces the number of timesteps during training, enabling SNN inference with just one timestep while maintaining high accuracy. The second focus of the thesis is on improving the visual understanding aspect of DNNs. The first work along this direction introduces a framework for visual syntactic understanding, drawing parallels between linguistic syntax and visual components of an image. By manipulating images to create syntactically incorrect examples and using a BERT-like autoencoder for reconstruction, the study significantly enhances the visual syntactic recognition capabilities of DNNs, evidenced by substantial improvements in classification accuracies on the CelebA and AFHQ datasets. Further, the thesis tackles unsupervised procedure learning from videos, given multiple videos of the same underlying task. Employing optimal transport (OT) and introducing novel regularization strategies, we develop the ‘OPEL’ framework, which substantially outperforms existing methods (27-46% average enhancement in F1-score) on both egocentric and third-person benchmarks. Overall, the dissertation advances the field by proposing brain-inspired models and novel learning frameworks that significantly enhance the efficiency and visual understanding capabilities of deep learning systems, making them more suitable for real-world applications.</p>