CIMeC - Center for Mind/Brain Sciences

Colloquium Series

The CIMeC Colloquium Series is an annual set of invited talks given by leading researchers in the mind/brain sciences, both from Italy and abroad, aimed principally at our PhD Students. Given the multi-disciplinary backgrounds of the CIMeC students and researchers, the colloquia are aimed at a general scientific level rather than at a more specialized audience. For school credit, all first, second and third year students attend the Colloquia. First and second year students prepare an essay based on one of the Colloquia of their choice summarizing the Colloquium, critically assessing the claims made and discussing the Colloquium in a broader context.

Find out who’s speaking at the next CIMeC Colloquium

Seminars 2019

CIMeC CLIC Seminar

Title: Bridging the image and text spaces with neural network methods for multimodal representation learning and spatial understanding
When: 21th November 2019, 3.15 - 4.15 p.m.
Where: Conference room, 1st floor, Palazzo Fedrigotti, Corso Bettini 31, Rovereto
Speaker: Guillem Collell,Language Intelligence and Information Retrieval group (LIIR) at KU Leuven
Abstract: Images and text often co-occur together in real world data. To perform real-world tasks, it is therefore crucial to learn good, actionable, distributed representations for each modality separately, as well as multimodal representations that concisely encode information from both modalities. In recent years, the emergence of Neural Networks and Deep Learning has eased this process and boosted performance in real-world tasks to a large extent. In this presentation, I discuss the use of feed-forward neural networks to map distributed distributed representations of one modality to the other, and in particular, I illustrate their use as a method to learn multimodal representations. Furthermore, I pinpoint some limitations of the feed-forward networks often employed to bridge modalities in applications such as cross-modal retrieval or zero-shot learning.
Spatial information is present in virtually every conceivable visual task, in one form or another. As such, spatial understanding has evolved as one of the main animal cognitive skills, being key to orientation, visual recognition and scene understanding tasks. As a result, human language is often grounded in spatial knowledge, being the spatial relationships between objects either explicitly specified with a spatial preposition (e.g., ``on" or ``below") or only implicitly implied through actions (e.g., ``riding"). Here, I approach the question of grounding spatial relations in text as a cross-modal mapping problem, and further illustrate the use of neural nets as a means of mapping the textual to the spatial (visual) domain, and vice versa. This enables visualizing and making explicit spatial relations otherwise only implicitly implied in text.

Train (Trentino Autism Initiative) Seminar

The seminar is co-organized by CIBIO and CIMeC
Title: Aging of immune system: causes, consequences and possible interventions
When: 17th July 2019 11.00 a.m.- 1.00 p.m.
Where: Aula B103 - Povo 2 (map)
Speaker: Luca Pangrazzi,  Institute for Biomedical Aging Research, University of Innsbruck
Abstract: Aging leads to a decline of immune function, a process known as immunosenescence, which contributes to higher incidence and severity of infectious diseases and decreased efficacy of vaccines in the elderly. Due to the involution of the thymus, the numbers of antigen-inexperienced (naïve) T cells are low in old age. For this reason, finding new strategies to support the maintenance of antigen-experienced adaptive immune cells generated during life is of utmost importance. In particular, the bone marrow (BM) has been shown to play the major role in the survival of memory cells.With our work, we first assessed how aging affects the maintenance of immunological memory in the human BM. A link was found between inflammation and oxidative stress, typical hallmarks of aging, and impaired survival of “healthy” adaptive immune cells. In the elderly, highly differentiated, senescent-like T cells, which migrate to the BM from the periphery, may compete for space with memory T cells. In addition, senescent-like T cells support BM inflammation through the production of pro-inflammatory molecules IFNγ and TNF. Because of their fundamental role played in this process, the phenotype of senescent T cell candidates was studied in details using microarrays. In parallel, strategies to specifically induce apoptosis in senescent T cells in vitro and to reduce BM inflammation and oxidative stress in vivo were identified. In summary, our work suggests that the maintenance of immunological memory in the elderly may be supported targeting directly the side effects of aging.

CIMeC CLIC Seminar

Title: Reasoning over meaning in context
When: 27thJune, 2019  10:00  - 11:30 a.m.
Where: Seminar Room, Third Floor, Palazzo Fedrigotti, Corso Bettini 31, Rovereto
Speaker: Katrin Erk, University of Texas at Austin
Abstract: How does meaning in context come about? Even when only two words are involved, the combination can be complex, as in the case of 'pet fish' (Osherson & Smith, 1981; Hampton 1985; Kamp & Partee, 1995), which is usually not eaten although fish often are, and is not fluffy although pets usually are. In a larger discourse, there are many interacting context influences, including selectional constraints, wider topic, and referents. If we view utterance interpretation as a process that generates a situation description (Fillmore, 1985), and if we view this process as probabilistic, then we can describe the different context influences as  interacting random variables. We take some first steps towards such a framework, and sketch how it could be used to analyze some examples.

Train (Trentino Autism Initiative) Seminar

The seminar is co-organized by DiPSCo and TRAIN
Title: Early recognition of autism spectrum disorder: from animal models to human studies
When: 4th June, 2019 3 p.m.
Where: Conference room, Palazzo Fedrigotti, 1st floor (Rovereto)
Speaker: Dr. Maria Luisa Scattoni (Istituto Superiore di Sanità, Roma)

Train (Trentino Autism Initiative) Seminar

The seminar is co-organized by the Center for Neuroscience and Cognitive Systems (CNCS) of IIT-UniTn and TRAIN (Trentino Autism Initiative)
Title: New therapeutic approaches to treat brain disorders characterized by impaired chloride homeostasis
When: 28th May , 2019 11 a.m.
Where: Conference room Palazzo Fedrigotti, 1st floor - corso Bettini 31, Rovereto
Speaker:  Dr. Laura Cancedda (IIT-Genova)

Train (Trentino Autism Initiative) Seminar

Title: Train seminar
When: 2nd April , 2019, 5.00 p.m.
Where: Sala Convegni Palazzo Fedrigotti, 1st floor (Rovereto)
Speaker: Prof. Martien Kas (University of Groningen, The Netherlands)
Abstract: Prof. Martien Kas (University of Groningen, The Netherlands) is an internationally known behavioural neuroscientist, with a long experience in genetic and animal model studies of neurodevelopmental disorders, including autism. He will give a lecture entitled "From Autism Spectrum Disorders to quantitative biology; a neurodevelopmental perspective”. In his lecture, he will give a a comprehensive overview on quantitative and neurodevelopmental studies to address the complexity of autism pectrum Disorders.

CIMeC CLIC Seminar

Title: Modeling reference games on real-world images: A common playground  for computational and experimental methods
When: 6th February 2019 2.00 P.M. - 3.30 p.m.
Where: Conference room, First Floor, Palazzo Fedrigotti, Corso Bettini 31, Rovereto
Speaker: Sina Zarrieß, University of Bielefeld
Abstract: For a long time, so-called reference games (cf. Rosenberg and Cohen, 1964) have been a popular experimental paradigm in linguistics, as they put together the basic ingredients of verbal interaction in a controlled microcosm: a speaker, a listener, a visual scene and a common communicative goal (identifying an object). Recently, this setting has been re-discovered in the Vision & Language community, and large-scale data sets that pair referring expressions with objects in real-world images have become available. These now constitute an excellent testbed for analyzing state-of-the-art machine learning models from a linguistic perspective and vice versa, and this is what I will do in this talk. First, I will present a computational study on neural referring expression generation, investigating different decoding strategies and how they relate to theoretical expectations. Second, I will present an ongoing eye-tracking study that leverages real-world images and a data-driven model for experimental hypothesis testing. Finally, I will discuss ideas on how to bring these two approaches even closer together.