Learning to Learn by Exploiting Prior Knowledge
One of the ultimate goals of open ended learning systems is to take advantage of experience to get a future benefit. We can identify two levels in learning. One builds directly over the data: it captures the pattern and regularities which allow for reliable predictions on new samples. The other starts from such an obtained source knowledge and focuses on how to generalize it to new target concepts: this is also known as learning to learn. Most of the existing machine learning methods stop at the first level and are able of reliable future decisions only if a large amount of training samples is available. This work is devoted to the second level of learning and focuses on how to transfer information from prior knowledge, exploiting it on a new learning problem with possibly scarce labeled data. We propose several algorithmic solutions by leveraging over prior models or features. One possibility is to constrain any target learning model to be close to the linear combination of several source models. Alternatively the prior knowledge can be used as an expert which judges over the target samples and considers the obtained output as an extra feature descriptor. All the proposed approaches evaluate automatically the relevance of prior knowledge and decide from where and how much to transfer without any need of external supervision or heuristically hand tuned parameters. A thorough experimental analysis shows the effectiveness of the defined methods both in case of interclass transfer and for adaptation across different domains. The last part of this work is dedicated to moving forward knowledge transfer towards life long learning. We show how to combine transfer and online learning to obtain a method which processes continuously new data guided by information acquired in the past. We also present an approach to exploit the large variety of existing visual data resources every time it is necessary to solve a new situated learning problem. We propose an image representation that decomposes orthogonally into a specific and a generic part. The last one can be used as an un-biased reference knowledge for future learning tasks.