Deep learning for computer vision: a comparison between convolutional neural networks and hierarchical temporal memories on object recognition tasks

In recent years, Deep Learning techniques have shown to perform well on a large variety of problems both in Computer Vision and Natural Language Processing, reaching and often surpassing the state of the art on many tasks. The rise of deep learning is also revolutionizing the entire field of Machine Learning and Pattern Recognition pushing forward the concepts of automatic feature extraction and unsupervised learning in general. However, despite the strong success both in science and business, deep learning has its own limitations. It is often questioned if such techniques are only some kind of brute-force statistical approaches and if they can only work in the context of High Performance Computing with tons of data. Another important question is whether they are really biologically inspired, as claimed in certain cases, and if they can scale well in terms of "intelligence". The dissertation is focused on trying to answer these key questions in the context of Computer Vision and, in particular, Object Recognition, a task that has been heavily revolutionized by recent advances in the field. Practically speaking, these answers are based on an exhaustive comparison between two, very different, deep learning techniques on the aforementioned task: Convolutional Neural Network (CNN) and Hierarchical Temporal memory (HTM). They stand for two different approaches and points of view within the big hat of deep learning and are the best choices to understand and point out strengths and weaknesses of each of them. CNN is considered one of the most classic and powerful supervised methods used today in machine learning and pattern recognition, especially in object recognition. CNNs are well received and accepted by the scientific community and are already deployed in large corporation like Google and Facebook for solving face recognition and image auto-tagging problems. HTM, on the other hand, is known as a new emerging paradigm and a new meanly-unsupervised method, that is more biologically inspired. It tries to gain more insights from the computational neuroscience community in order to incorporate concepts like time, context and attention during the learning process which are typical of the human brain. In the end, the thesis is supposed to prove that in certain cases, with a lower quantity of data, HTM can outperform CNN.

Active Translation into English and Italian of Hosseini's adaptation of the Persian epic,"Shahname"

This thesis proposes a translation from Persian into Italian and English of an ancient Persian epic called Shahname, or literally “The Book of Kings,” by Ferdosi, first published in the 11th century CE. The translation proposed, however, is not based on the original book by Ferdosi, which is written all in verse, but rather, an edited, shorter, and simplified version written in prose, by Mohamad Hosseini, first published in 2013. Nonetheless, in his version, Hosseini included some of the verses from the original poems in order to show the value and the beauty of Ferdosi’s writing. Many translations of Ferdosi’s book have been made into English, but only one translation has been made into Italian, by one Italo Pizzi, in 8 volumes, all in verse, in 1886. This thesis analyses and discusses the choices made for the two translations presented into English and Italian. My project is not only to propose translations of Hosseini’s version, but to also introduce the reader to the Persian culture, and to the life of the most famous Iranian epic writer, Ferdosi, and his masterpiece, Shahname.

Trauma-induced coagulopathy in severely injured patients: knowledge lost in translation?

Many guidelines exist on how to treat patients with multiple injuries correctly in an accident and emergency setting. The aim of the present work was to find out how well patients are treated focusing on trauma induced coagulopathy (TIC), and what anaesthetists involved in trauma care think about their own experiences with TIC.

Lost in translation: the impact of publication language on citation frequency in the scientific dental literature

Citation metrics are commonly used as a proxy for scientific merit and relevance. Papers published in English, however, may exhibit a higher citation frequency than research articles published in other languages, though this issue has not yet been investigated from a Swiss perspective where English is not the native language.

Protein disulfide isomerase blocks CEBPA translation and is up-regulated during the unfolded protein response in AML

Deregulation of the myeloid key transcription factor CEBPA is a common event in acute myeloid leukemia (AML). We previously reported that the chaperone calreticulin is activated in subgroups of AML patients and that calreticulin binds to the stem loop region of the CEBPA mRNA, thereby blocking CEBPA translation. In this study, we screened for additional CEBPA mRNA binding proteins and we identified protein disulfide isomerase (PDI), an endoplasmic reticulum (ER) resident protein, to bind to the CEBPA mRNA stem loop region. We found that forced PDI expression in myeloid leukemic cells in fact blocked CEBPA translation, but not transcription, whereas abolishing PDI function restored CEBPA protein. In addition, PDI protein displayed direct physical interaction with calreticulin. Induction of ER stress in leukemic HL60 and U937 cells activated PDI expression, thereby decreasing CEBPA protein levels. Finally, leukemic cells from 25.4% of all AML patients displayed activation of the unfolded protein response as a marker for ER stress, and these patients also expressed significantly higher PDI levels. Our results indicate a novel role of PDI as a member of the ER stress-associated complex mediating blocked CEBPA translation and thereby suppressing myeloid differentiation in AML patients with activated unfolded protein response (UPR).

The persistence of musical memories: A descriptive study of earworms

We describe some characteristics of persistent musical and verbal retrieval episodes, commonly known as "earworms." In Study 1, participants first filled out a survey summarizing their earworm experiences retrospectively. This was followed by a diary study to document each experience as it happened. Study 2 was an extension of the diary study with a larger sample and a focus on triggering events. Consistent with popular belief, these persistent musical memories were common across people and occurred frequently for most respondents, and were often linked to recent exposure to preferred music. Contrary to popular belief, the large majority of such experiences were not unpleasant. Verbal earworms were uncommon. These memory experiences provide an interesting example of extended memory retrieval for music in a naturalistic situation.