Attention, transformer & Chat-GPT: The second galilean revolution in the making? What the hype is about.
The introduction of the attention mechanism 2014/17 initiated an AI revolution. Instead of classical deep neural networks, large language models are based nowadays on 'transformer' architectures implementing the "attention is all you need" paradigm. In this lecture an overview of state-of-the-art machine learning concepts is presented. The aim is an understanding of 'Generative Pretrained Transformers', the basis of ChatGPT. The possible functional similarity of active information routing in foundation models and in the brain is discussed. How likely is the advent of the second Galilean revolution, in this view, namely that the human brain will lose its hitherto unique position?
local host: Prof. Dr. Reinhard Dörner | doerner@atom.uni-frankfurt.de
With the Ukraine War, the threat of nuclear weapons use has re-emerged in the public debate. Today, there exist over 12,000 nuclear weapons globally, and large fissile material stocks allowing the production of many more. While the war and the political climate will not allow disarmament initiatives in the foreseeable future, the public debate has triggered a new sense of urgency. Physicists have an important role: To enable international agreements on warhead and fissile material reductions, strong verification protocols are essential to monitor compliance. New concepts and techniques will be required and must be available should a political window of opportunity open in the longer term. As they can take many years to develop, continuing this work remains crucial today. In this presentation, three elements of a possible verification toolbox will be presented. First, radiation detection techniques suited to establish the authenticity of nuclear warheads to be dismantled must be developed. Second, new concepts are required to estimate the amount of produced weapons-usable fissile materials. This is the purpose of nuclear archaeology, which attempts to reconstruct the past fissile material production using forensic measurements in shut-down nuclear facilities and extensive simulations. Third, we discuss the detection of antineutrinos from nuclear waste as part of nuclear archaeology.
local host: Fachschaft Physik | kontakt@fachschaft.physik-uni-frankfurt.de
Cosmology
has evolved from a speculative field into a precision science using large sky
surveys as a laboratory for fundamental physics. From this year, the Euclid
satellite and Rubin Observatory will map the distribution of galaxies across
most of the sky and 10 billion years of cosmic history. They will identify
billions of galaxies, estimate their distances and determine their shapes to
test physics on the largest scales.
In this
talk, I will take you on a journey through our Universe across the largest scales
and the longest times. We will discover the large-scale skeleton of matter
behind the cosmic web of structure observed in galaxy surveys. I will explain
how our Universe evolved from a nearly uniform initial state into today's
cosmos with rich structure from stars to galaxies and beyond. We will see how
the tug of war between the gravity of dark matter and the expansion by dark
energy is recorded in the cosmic large-scale structure. I will describe how we
can squeeze out more information from the largest galaxy surveys by probing
cosmic structures beyond the average of standard forward models and statistical
analyses. This will enable us to put our physical model of the Universe and its
ingredients to the ultimate test.
local host: Prof. Dr. Dirk Rischke | drischke@itp.uni-frankfurt.de
The next generation of ground based gravitational wave observatories
The current
"advanced" generation of gravitational wave detectors has completed
an impressive series of observational runs in recent years, and will have
started another data run by the time of the talk. However, the sensitivity of
today's detectors, impressive as it is, only allows us to eavesdrop on sources
in our cosmic neighbourhood, and not yet with sufficient precision to study the
underlying physical processes in detail.
With the Einstein telescope and the Cosmic
Explorer, the next generation of gravitational wave detectors is being planned.
These detectors will be an order of magnitude more sensitive and extend to
lower frequencies, opening up new areas and looking back to the early times
after the Big Bang. This talk will give an overview of the possibilities, the
plans and the difficulties to achieve this.
.
local host: Prof. Dr. Luciano Rezzolla | rezzolla@itp.uni-frankfurt.de
Physik
erklären - was zeichnet gutes Erklären aus?
Erklären zu können, wird häufig von Schülerinnen und Schülern als zentrale Eigenschaft guter Physiklehrkräfte genannt. Aber was bedeutet es eigentlich, gut erklären zu können? Was macht also eine Erklärung mehr oder weniger verständlich? Dürfen Lehrkräfte in einem zeitgemäßen Physikunterricht überhaupt erklären oder ist es lernwirksamer, wenn sich Schülerinnen und Schüler die Physik selbst erarbeiten, z.B. anhand von Schülerexperimenten? Wie lernwirksam können dann Erklärvideos, z.B. bei YouTube, sein? Welche Fähigkeiten muss jemand haben, um gut erklären zu können - und wie fördert man die im Lehramtsstudium? Im Vortrag werden mehrere zusammenhängende empirische Studien diskutiert, die auf diese Fragen eingehen.
local host: Prof. Dr. Thomas Wilhelm | wilhelm@physik.uni-frankfurt.de
14.06.2023
PD Dr. Markus Röllig (Physikalischer Verein) - Antrittsvorlesung
tba
local host: Prof. Dr. René Reifarth | reifarth@physik.uni-frankfurt.de
21.06.2023
Dr. Dirk Eidemüller (Wissenschaftspublizist)
Nuclear Energy
local host: Prof. Dr. Luciano Rezzolla | rezzolla@itp.uni-frankfurt.de
28.06.2023
Festkolloquium Herbert Ströbele (Institut für Kernphysik, Goethe-Universität Frankfurt)
tba
local host: Prof. Dr. Henner Büsching | buschin@ikf.uni-frankfurt.de
05.07.2023
Prof. Dr. Christian Fischer (Justus-Liebig-Universität Gießen)
tba
local host: Prof. Dr. Dirk Rischke | drischke@itp.uni-frankfurt.de
12.07.2023
Prof. Dr. Mathias Kläui (Universität Mainz)
Antiferromagnetic
Spin- and Orbitronics
local host: Prof. Dr. Cornelius Krellner | Krellner@Physik.uni-frankfurt.de
