Attention mechanisms for semantic few-shot learning

Download
2021-9-1
Baran, Orhun Buğra
One of the fundamental difficulties in contemporary supervised learning approaches is the dependency on labelled examples. Most state-of-the-art deep architectures, in particular, tend to perform poorly in the absence of large-scale annotated training sets. In many practical problems, however, it is not feasible to construct sufficiently large training sets, especially in problems involving sensitive information or consisting of a large set of fine-grained classes. One of the main topics in machine learning research that aims to address such limitations is few-shot learning where only few labeled samples are made available for each novel class of interest. An inherent difficulty in few-shot learning is the various ambiguities resulting from having only few training samples per class. To tackle this fundamental challenge in few-shot learning, in this thesis, we propose an approach that aims to guide the meta-learner via semantic priors. To this end, we build meta-learning models that can benefit from prior knowledge based semantic representations of classes of interest when synthesizing target classifiers. We propose semantically-conditioned feature attention and sample attention mechanisms that estimate and utilize the importance of representation dimensions and training instances. In sample attention, we aim to weigh each individual training example based on its representativeness for the related class. We, then, use the information extracted from each example proportional to its individual weight. In feature attention, we aim to weigh each visual feature dimension based on the semantic embedding vectors we obtain for each class. We also study the problem of sample noise in few-shot learning, where some training examples are irrelevant due to annotation or data collection errors, which can be the case for various real-world problems. Our experimental results demonstrate the effectiveness of the proposed semantic few-shot learning model with and without sample noise.

Suggestions

Semantics-driven attentive few-shot learning over clean and noisy samples
Baran, Orhun Buğra; Cinbiş, Ramazan Gökberk (2022-11-01)
Over the last couple of years, few-shot learning (FSL) has attracted significant attention towards minimiz-ing the dependency on labeled training examples. An inherent difficulty in FSL is handling ambiguities resulting from having too few training samples per class. To tackle this fundamental challenge in FSL, we aim to train meta-learner models that can leverage prior semantic knowledge about novel classes to guide the classifier synthesis process. In particular, we propose semantically-conditioned featur...
Recursive Compositional Reinforcement Learning for Continuous Control Sürekli Kontrol Uygulamalari için Özyinelemeli Bileşimsel Pekiştirmeli Öǧrenme
Tanik, Guven Orkun; Ertekin Bolelli, Şeyda (2022-01-01)
Compositional and temporal abstraction is the key to improving learning and planning in reinforcement learning. Modern real-world control problems call for continuous control domains and robust, sample efficient and explainable control frameworks. We are presenting a framework for recursively composing control skills to solve compositional and progressively complex tasks. The framework promotes reuse of skills, and as a result quickly adaptable to new tasks. The decision-tree can be observed, providing insi...
Closed-form sample probing for training generative models in zero-shot learning
Çetin, Samet; Cinbiş, Ramazan Gökberk; Department of Computer Engineering (2022-2-10)
Generative modeling based approaches have led to significant advances in generalized zero-shot learning over the past few-years. These approaches typically aim to learn a conditional generator that synthesizes training samples of classes conditioned on class embeddings, such as attribute based class definitions. The final zero-shot learning model can then be obtained by training a supervised classification model over the real and/or synthesized training samples of seen and unseen classes, combined. Therefor...
Gradient Matching Generative Networks for Zero-Shot Learning
Sariyildiz, Mert Bulent; Cinbiş, Ramazan Gökberk (2019-01-01)
Zero-shot learning (ZSL) is one of the most promising problems where substantial progress can potentially be achieved through unsupervised learning, due to distributional differences between supervised and zero-shot classes. For this reason, several works investigate the incorporation of discriminative domain adaptation techniques into ZSL, which, however, lead to modest improvements in ZSL accuracy. In contrast, we propose a generative model that can naturally learn from unsupervised examples, and synthesi...
Design of a problem-based online learning environment and evaluation of its effectiveness
Gündüz, Abdullah Yasin; Alemdağ, Ecenaz; Yaşar, Sevil; Erdem, Mukaddes (2017-07-01)
Problem-based learning approach present several advantages such as improving students’ engagement in learning and fostering their higher-order thinking skills. Although there is a plethora of research regarding implementation of problem-based learning in classrooms, its design and application process for web-based environments need further investigation because of independent nature of online settings. This study developed a problem-based online learning environment based on constructivist learning design m...
Citation Formats
O. B. Baran, “Attention mechanisms for semantic few-shot learning,” M.S. - Master of Science, Middle East Technical University, 2021.