Seminaria ogólnoinstytutowe:

Streszczenie (autorskie):

Podczas seminarium zaprezentujemy rodzinę modeli językowych PLLuM opracowaną na zlecenie Ministerstwa Cyfryzacji przez polskie ośrodki badawcze z myślą o specyfice języka polskiego, lokalnym kontekście kulturowym oraz potrzebach krajowej administracji publicznej. W trakcie wystąpnienia omówimy główne założenia i cele projektu PLLuM, a także proces tworzenia modeli. Szczególną uwagę poświęcimy pracom realizowanym przez IPI PAN obejmującym pozyskiwanie i opracowanie danych tekstowych, budowę korpusów instrukcji i preferencji, wychowanie modeli oraz ich ewaluację, w tym ewaluację zdolności generatywnych oraz zastosowanych mechanizmów zabezpieczeń.

Streszczenie (autorskie):

The current huge successes of neural networks and deep learning sidelined the important aspect of transparency of the predictions made by these classifiers. While predictive accuracy is important in many applications, in the case of knowledge discovery in life sciences the main interest is in understanding how the predictions are made and thus some loss of accuracy may be accepted. In this seminar, I shall present our methodology based on the combination of Monte Carlo Feature Selection (rmcfs) and rough set R.ROSETTA developed in my laboratory that supports making discoveries in life sciences, but is general enough to contribute to predictive maintenance. Examples will include transcriptomic studies of Systemic Lupus Erythematosus, a complex immune system disease, progression of Acute Myeloid Leukemia and establishment of protection against SIV/HIV in Rhesus macaques upon vaccination. And for the predictive maintenance case, I shall use the model of water piping system in the Stockholm area developed in our laboratory.

Streszczenie (autorskie):

In this talk I will present a general introduction to topological data analysis suitable for the use in practical statistics. There are two aspects in which topology can contribute to statistical learning; in data visualization, for exploratory data analysis and for feature extraction, to make the existing statistical techniques potentially more efficient in practice. We will start from data visualization methods and present a number of topological visualization tools, including mapper, ball mapper and ClusterGraph. Subsequently we will discuss some basic data features, focusing on concepts originating from Euler characteristics. We will show how it can be utilized for topological goodness-of-fit tests as well as as a tool to extract information from various branches of sciences, including image analysis and dynamical systems.

Streszczenie (autorskie):

Daniel Dennett speculated in Kinds of Minds (1996):
“Perhaps the kind of mind you get when you add language to it is so different from the kind of mind you can have without language that calling them both minds is a mistake.”

Recent work in AI can be seen as testing Dennett’s thesis by exploring the performance of AI systems with and without linguistic training. I will argue that the success of Large Language Models at inferential reasoning, limited though it is, supports Dennett’s radical view about the effect of language on thought. I suggest it is the abstractness and efficiency of linguistic encoding that lies behind the capacity of LLMs to perform inferences across a wide range of domains. In a slogan, language makes inference computationally tractable. I will assess what these results in AI indicate about the role of language in the workings of our own biological minds.