BEGIN:VCALENDAR
VERSION:2.0
PRODID:icalendar-ruby
CALSCALE:GREGORIAN
METHOD:PUBLISH
BEGIN:VTIMEZONE
TZID:Europe/Vienna
BEGIN:DAYLIGHT
DTSTART:20260329T030000
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=3
TZNAME:CEST
END:DAYLIGHT
BEGIN:STANDARD
DTSTART:20261025T020000
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=10
TZNAME:CET
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260615T191843Z
UID:1773738000@ist.ac.at
DTSTART:20260317T100000
DTEND:20260317T110000
DESCRIPTION:Speaker: Mirza Baig\nhosted by Julia Reisenbauer\nAbstract: Blo
 ckchains enable distributed consensus in permissionless settings\, where p
 articipants are unknown\, dynamically changing\, and do not trust each oth
 er. While Bitcoin\, based on Proof-of-Work (PoW)\, was the first protocol 
 in this model\, significant research has focused on permissionless protoco
 ls using alternative physical resources\, specifically Proof-of-Space (PoS
 pace) and Verifiable Delay Functions (VDFs). This thesis investigates the 
 theoretical limits and design space of longest-chain protocols in the full
 y permissionless and dynamically available settings using these three reso
 urces. First\, we address the feasibility of blockchains relying solely on
  storage as a resource. We prove a fundamental impossibility result: there
  exists no secure longest-chain protocol based exclusively on Proof-of-Spa
 ce in the fully permissionless or dynamically available settings. Further\
 , we quantify the adversarial capabilities required to execute a double-sp
 end attack. Our result formally justifies the necessity of coupling PoSpac
 e with time-dependent primitives (such as VDFs) or to move to less permiss
 ive settings (quasi-permissionless or permissioned) to ensure security.Sec
 ond\, we generalize Nakamoto-like heaviest chain consensus to protocols ut
 ilizing combinations of multiple physical resources. We analyze chain sele
 ction rules governed by a weight function Γ(S\, V\,W)\, which assigns wei
 ght to blocks based on recorded Space (S)\, VDF speed (V )\, and Work (W).
  We provide a complete classification of secure weight functions\, proving
  that a weight function is secure against private double-spend attacks if 
 and only if it is homogeneous in the timed resources (V\,W) and sub-homoge
 neous in S. This framework unifies existing protocols like Bitcoin and Chi
 a under a single theoretical model and provides a powerful tool for design
 ing new longest-chain blockchains from a mix of physical resources.
LOCATION:Moonstone Bldg / Ground floor / Seminar Room C (I24.EG.030c) and Z
 oom\, ISTA
ORGANIZER:
SUMMARY:Mirza Baig: Thesis Defense: On Secure Chain Selection Rules from Ph
 ysical Resources in a Permissionless Setting
URL:https://talks-calendar.ista.ac.at/events/6331
END:VEVENT
BEGIN:VEVENT
DTSTAMP:20260615T191843Z
UID:1774011600@ist.ac.at
DTSTART:20260320T140000
DTEND:20260320T150000
DESCRIPTION:Speaker: Tobias Kleinhanns\nhosted by Jorryt Matthee\nAbstract:
  Thermoelectric (TE) technology directly converts between heat and electri
 city\, enabling both waste heat harvesting and localized cooling applicati
 ons. Yet\, TE technology remains limited to niche applications\, mostly du
 e to high fabrication costs\, and lower efficiencies compared to other ene
 rgy conversion technologies. Solution processing provides a scalable and c
 heaper route than conventional solid-state methods\, potentially enabling 
 the broader adoption of TE technology. Beyond its cost benefits\, solution
  processing uniquely allows the synthesis of well-defined nanocrystal buil
 ding blocks\, translating nanoscale defect control into bulk structures. S
 uch microstructure engineering across length scales allows rational optimi
 zation of the TE performance. This thesis explores defect engineering of s
 olution-processed silver selenide (Ag2Se)\, with focus on compositional an
 d microstructural control for enhanced TE performance\, specifically throu
 gh thiol-amine based synthesis and post-synthetic treatment using polyanio
 nic CdSe complexes. The material library was extended to lead selenide (Pb
 Se)\, whose TE performance was optimized through chlorine (Cl) doping from
  functionalized two-dimensional titanium carbides (MXenes).
LOCATION:Sunstone Bldg / Ground floor / Big Seminar Room B (I23.EG.102) \, 
 ISTA
ORGANIZER:
SUMMARY:Tobias Kleinhanns: Thesis Defense: Unraveling the Origin and Evolut
 ion of Defects to Enable Advanced Thermoelectric Performance
URL:https://talks-calendar.ista.ac.at/events/6341
END:VEVENT
BEGIN:VEVENT
DTSTAMP:20260615T191843Z
UID:1776081600@ist.ac.at
DTSTART:20260413T140000
DTEND:20260413T150000
DESCRIPTION:Speaker: Ksenia Khudiakova\nhosted by Mikhail Lemeshko\nAbstrac
 t: This Ph.D. thesis investigates how different forms of selection shape g
 enetic diversity in a constant environment that has led to evolutionary eq
 uilibrium. We progress from a non-epistatic case to increasingly complex m
 odels of epistasis and rely on stochastic and deterministic theory togethe
 r with simulations.We first show that in the non-epistatic case\, and in a
 n asexual population\, weak purifying selection leads to multiple-merger g
 enealogies\, and that the transition in genealogical properties coincides 
 with the onset of Muller’s ratchet.Then\, we apply a result from discret
 e Morse theory to prove that adding each next fitness peak to the landscap
 e requires at least one additional special pairwise interaction between lo
 ci\, called reciprocal sign epistasis.We then show that reciprocal sign ep
 istasis can extend the diversity-promoting effects of balancing selection 
 and delineate the parameter conditions under which this effect is expected
  to occur.Finally\, we demonstrate how epistasis that arises under stabili
 zing selection amplifies the effects of random genetic drift by causing th
 e selection coefficients of mutations to fluctuate through interactions wi
 th the changing genetic background.Together\, these results show that non-
 epistatic purifying selection reduces genetic diversity relative to neutra
 lity\, and that once Muller’s ratchet starts operating\, this reduction 
 cannot be captured by any simple rescaling of Kingman’s coalescent. On a
 n epistatic fitness landscape\, reciprocal sign epistasis is a key ingredi
 ent for generating multiple fitness peaks\, and it substantially alters wi
 thin-population dynamics at evolutionary equilibrium by extending the dive
 rsity-maintaining effects of balancing selection and driving temporal chan
 ges in selection coefficients. This Ph.D. thesis thus advances our underst
 anding of how epistasis shapes genetic diversity and evolutionary dynamics
  in populations at equilibrium.
LOCATION:Sunstone Bldg / Ground floor / Big Seminar Room B (I23.EG.102) and
  Zoom\, ISTA
ORGANIZER:
SUMMARY:Ksenia Khudiakova: Thesis Defense: How epistasis and purifying sele
 ction shape genetic diversity
URL:https://talks-calendar.ista.ac.at/events/6399
END:VEVENT
BEGIN:VEVENT
DTSTAMP:20260615T191843Z
UID:1776672000@ist.ac.at
DTSTART:20260420T100000
DTEND:20260420T110000
DESCRIPTION:Speaker: Eugenia Iofinova\nhosted by Samara Ren\nAbstract: As n
 eural-network-based models grow both in size and popularity\, interest has
  grown in making the models smaller and more efficient to train. To that e
 nd\, many methods have been proposed to prune models by reducing their num
 ber of nonzero parameters. Additionally\, parameter-efficient fine-tuning\
 , in which a much smaller number of parameters than the total contained in
  the model is updated during training\, has become very popular\, especial
 ly in the space of Large Language Models. At the same time\, the increasin
 gly routine deployment of machine learning in real-world applications has 
 spurred a drive to make them more trustworthy - in the sense of\, among ot
 her things\, being unbiased\, interpretable\, and editable. In this thesis
 \, we examine the interplay between efficiency and trustworthiness.First\,
  we analyze the effects of model pruning on bias in computer vision models
 \, demonstrating that increased sparsity leads to greater bias\, largely a
 s a function of increased model uncertainty in marginal cases. Based on th
 is observation\, we propose several bias mitigation techniques. Then\, we 
 demonstrate that example-specific model pruning can improve model interpre
 tation methods while improving pruning efficiency to make example-specific
  model pruning feasible in real time. Then\, we investigate the effectiven
 ess of parameter-efficient and data-efficient model personalization via fi
 ne-tuning\, demonstrating that it is highly feasible with very small compu
 tational and data resources. Finally\, we consider efficiency in editing m
 odel knowledge using a custom synthetic data framework\, demonstrating tha
 t parameter-efficient\, low-rank fine-tuning frequently outperforms full-r
 ank fine-tuning\, and\, additionally\, restricting fine-tuning to specific
  model blocks frequently improves results. Together\, the results in this 
 thesis provide new insights and techniques for combining trustworthiness a
 nd efficiency during neural network inference and training.
LOCATION:Office Bldg West / Ground floor / Heinzel Seminar Room (I21.EG.101
 ) and Zoom\, ISTA
ORGANIZER:
SUMMARY:Eugenia Iofinova: Thesis Defense: On the Utility and Effects of Eff
 iciency in Artificial Neural Networks
URL:https://talks-calendar.ista.ac.at/events/6391
END:VEVENT
BEGIN:VEVENT
DTSTAMP:20260615T191843Z
UID:1778065200@ist.ac.at
DTSTART:20260506T130000
DTEND:20260506T160000
DESCRIPTION:Speaker: David Vijatovic\nhosted by Amelia Douglass\nAbstract: 
 Motor circuits must generate distinct patterns of movement while adapting 
 to developmental changes in body form and behavioral demands. Frog metamor
 phosis provides a unique opportunity to study this process\, as Xenopus la
 evis transitions from axial\, tail-based swimming to limb-based locomotion
 . This work investigates how spinal circuits are molecularly organized and
  functionally reorganized during this transition.I examine how spinal neur
 on composition changes across metamorphosis\, showing that cell type diver
 sity increases with the emergence of limb movement. In particular\, inhibi
 tory V1 interneurons and motor neurons expand and diversify into transcrip
 tionally defined populations resembling those described in mammals. These 
 findings support conserved organizational principles of vertebrate spinal 
 circuits and link cell type diversification to the emergence of more compl
 ex motor behaviors. I further develop viral and optical approaches to prob
 e how these molecularly defined circuits are functionally reorganized duri
 ng this transition.Together\, these findings establish Xenopus laevis as a
  powerful model for linking molecular identity\, circuit dynamics\, and be
 havior across vertebrate development.
LOCATION:Sunstone Bldg / Ground floor / Big Seminar Room A / 27 seats (I23.
 EG.102)\, ISTA
ORGANIZER:
SUMMARY:David Vijatovic: Thesis Defense:Dissecting molecular and functional
  basis of motor control in Xenopus laevis frog
URL:https://talks-calendar.ista.ac.at/events/6442
END:VEVENT
BEGIN:VEVENT
DTSTAMP:20260615T191843Z
UID:1779458400@ist.ac.at
DTSTART:20260522T160000
DTEND:20260522T170000
DESCRIPTION:Speaker: Jeferson Zapata\nhosted by Matthew Kwan\nAbstract: Sta
 ndard mathematical optimization and root-finding algorithms often stall or
  fail when faced with degenerate or singular problems. This thesis present
 s novel symbolic-numeric frameworks designed to overcome these theoretical
  and computational bottlenecks. The first part of the presentation introdu
 ces a hybrid algorithm that reformulates weakly feasible\, degenerate semi
 definite programs (SDPs) into structured polynomial systems\, enabling the
  rigorous algebraic certification of numerical approximations. The second 
 part addresses the loss of convergence in path-tracking methods near isola
 ted singular roots. By modeling solution paths as generalized fractional P
 uiseux series and utilizing an explicitly derived algebraic predictor\, th
 e proposed algorithms restore superlinear convergence and significantly re
 duce computational overhead in heavily degenerate systems.
LOCATION:Office Bldg West / Ground floor / Heinzel Seminar Room (I21.EG.101
 ) and Zoom\, ISTA
ORGANIZER:
SUMMARY:Jeferson Zapata: Thesis Defense: Overcoming Degeneracy and Singular
 ity: Techniques for Semidefinite Programs and Homotopy Continuation Endgam
 es
URL:https://talks-calendar.ista.ac.at/events/6476
END:VEVENT
BEGIN:VEVENT
DTSTAMP:20260615T191843Z
UID:1780488000@ist.ac.at
DTSTART:20260603T140000
DTEND:20260603T150000
DESCRIPTION:Speaker: Antoine El-Hayek\nhosted by Georgios Katsaros\nAbstrac
 t: Networks can change over time. Whether it is the Facebook network\, whe
 re 'friendships' emerge or end\, the internet\, where computers connect or
  disconnect from the network\, or the train network\, where new lines can 
 be created or canceled. This creates specific challenges for algorithms de
 signed to either run or manage these 'dynamic' networks.A distributed algo
 rithm is an algorithm that runs on the network. Think of the internet: for
  computers to communicate through this network\, the algorithm should be i
 nstalled on both computers\, and both computers won't necessarily have acc
 ess to the same information. If the network is unstable\, or is simply dyn
 amic\, the communication between them can drop at any point\, and the algo
 rithm should be robust to such changes. In this model\, we look at two pro
 blems: the broadcast problem\, where one computer tries to send a message 
 to every single other computer on the network\, and the majority problem\,
  where initially each computer is given an opinion among k possible ones\,
  and the goal is for every computer to know which opinion was originally t
 he most popular one.A dynamic graph algorithm is an algorithm that manages
  a dynamic network\, or computes a property on it. Think of the network of
  webpages. You might want to rank all webpages from the most important one
  to the least (popular search engines call it the 'PageRank' algorithm). H
 owever\, every day\, new pages go online\, while old pages go offline. You
  wouldn't want the algorithm to be run from scratch every night over all o
 nline webpages\, as this would cost too much energy. Instead\, you would l
 ike to have an algorithm that\, once it has computed the rank of each webp
 age at a given time\, be able to handle a few changes\, and update its sol
 ution accordingly. In the dynamic graph algorithm model\, we look at the m
 inimum cut problem\, where the goal is is divide the network into two subn
 etworks\, such that the number of connections from one subnetwork to the o
 ther is minimized.
LOCATION:Moonstone Bldg / Ground floor / Seminar Room F (I24.EG.030f) and Z
 oom\, ISTA
ORGANIZER:
SUMMARY:Antoine El-Hayek: Thesis Defense: Handling Updates and Failures: Dy
 namic Graph Algorithms and Distributed Computing on Dynamic Networks
URL:https://talks-calendar.ista.ac.at/events/6452
END:VEVENT
BEGIN:VEVENT
DTSTAMP:20260615T191843Z
UID:1781607600@ist.ac.at
DTSTART:20260616T130000
DTEND:20260616T140000
DESCRIPTION:Speaker: Konstantin Kueffner\nhosted by Krzysztof Pietrzak\nAbs
 tract: As automated decision-makers have become ubiquitous in many domains
  of life\, their decisions have become increasingly consequential. Recent 
 years have shown that such systems can exhibit discriminatory behaviour ag
 ainst individuals and social groups alike\, thereby amplifying existing bi
 ases and entrenching socio-economic disparities over time. Algorithmic fai
 rness addresses this problem by developing methods to quantify and mitigat
 e unfair behaviour. However\, much of the existing literature studies fair
 ness in a static pre-deployment setting and\, therefore\, neglects that au
 tomated decision-makers are often deployed in dynamic environments\, where
  their behaviour and the populations they affect may change over time.This
  thesis addresses this gap through the lens of runtime verification. Inste
 ad of treating fairness as a property of a classifier together with a fixe
 d input distribution\, it reframes fairness as a property of the interacti
 on trace between the decision-maker and its deployment environment. To eva
 luate such sequential fairness properties\, the thesis develops runtime mo
 nitors that observe the evolving interaction between the system and the en
 vironment and issue verdicts after each new observation. Because\, these m
 onitors are designed to detect unfair behaviour during deployment\, they c
 omplement fair training\, auditing\, verification\, and enforcement by pr
 oviding an additional layer of mathematically rigorous fairness assurance.
 In summary\, the thesis develops quantitative\, trace-based analogues of c
 lassical group and individual fairness measures and constructs monitors fo
 r them. This includes monitors for long-run group fairness over Markovian 
 traces\, for the time-varying welfare of a changing population in a dynami
 cal system\, and for the individual fairness of an arbitrary system genera
 ting a trace of inputs and outputs. To achieve this\, the monitors combine
  ideas from runtime verification\, sequential statistics\, and nearest-nei
 ghbour search. In the group-fairness settings\, monitoring is primarily a 
 sequential statistical estimation problem: the monitor must construct stat
 istically sound interval estimates of fairness values from dependent and p
 artially observed interactions. In the individual-fairness setting\, the m
 ain challenge is computational efficiency: the monitor must detect individ
 ual fairness violations by efficiently comparing the current decision with
  all previously observed decisions.
LOCATION:Central Bldg / O1 / Mondi 3 (I01.O1.010) and Zoom\, ISTA
ORGANIZER:
SUMMARY:Konstantin Kueffner: Thesis Defense: Monitoring Algorithmic Fairnes
 s in Sequential Decision Making
URL:https://talks-calendar.ista.ac.at/events/6480
END:VEVENT
BEGIN:VEVENT
DTSTAMP:20260615T191843Z
UID:1781703000@ist.ac.at
DTSTART:20260617T153000
DTEND:20260617T163000
DESCRIPTION:Speaker: Yunzhe Li\nhosted by Maksym Serbyn\nAbstract: We study
  spectral rigidity and nonrigidity phenomena in dynamical systems. The cen
 tral question is whether a dynamical system can be determined\, up to a na
 tural conjugacy\, by its spectrum.The first part of the talk focuses on st
 andard maps from the viewpoint of action spectra. We construct nontrivial 
 deformations of the standard map that preserve the symplectic actions of i
 nfinitely many periodic orbits accumulating onto an invariant curve. This 
 result can be viewed as a symplectic twist-map analogue of length-spectral
  nonrigidity phenomena for Riemannian manifolds and convex billiards\, mot
 ivating the problem of constructing analogous “partially length-isospect
 ral” deformations of strictly convex billiard tables. The proof combines
  a resonant normal form construction with Picard iteration schemes to prod
 uce a sequence of periodic orbits accumulating on an invariant curve with 
 a Liouville rotation number. The second part of the talk briefly explores
  rigidity questions for Liouville metrics on the two-dimensional torus. A 
 long-standing folklore conjecture asserts that Liouville metrics are the o
 nly integrable metrics on the torus. We give a length-spectral rigidity re
 sult for the class of trigonometric conformal deformations of Liouville me
 trics by exploiting the dynamical properties of rational tori\, which are 
 analogues of resonant convex caustics in billiards. We also establish a co
 mplementary classification result showing that marked-length-isospectral L
 iouville metrics are characterized by rearrangements of the one-dimensiona
 l functions appearing in their conformal factors\, generalizing a theorem 
 of Abbondandolo and Mazzucchelli. In particular\, this result yields many 
 nonrigidity examples within the class of Liouville metrics.
LOCATION:Office Bldg West / Ground floor / Heinzel Seminar Room (I21.EG.101
 ) and Zoom\, ISTA
ORGANIZER:
SUMMARY:Yunzhe Li: Thesis Defense: Spectral Rigidity and Nonrigidity of Dyn
 amical Systems
URL:https://talks-calendar.ista.ac.at/events/6506
END:VEVENT
BEGIN:VEVENT
DTSTAMP:20260615T191843Z
UID:1782201600@ist.ac.at
DTSTART:20260623T100000
DTEND:20260623T110000
DESCRIPTION:Speaker: Lea Becker\nhosted by Robert Seiringer\nAbstract: Char
 acterizing protein dynamics at the atomic level is essential for our under
 standing of biological mechanisms. Whether it is to facilitate metabolite 
 transport\, catalyze reactions\, transmit signals\, or regulate metabolism
  – proteins are constantly in motion and sample multiple conformational 
 states to fulfill their function. Nuclear magnetic resonance (NMR) spectro
 scopy is particularly well suited to elucidate the dynamics of biomolecule
 s on their complex free-energy landscape. In particular\, solid-state magi
 c-angle spinning (MAS) NMR enables the study of large molecular assemblies
 \, protein crystals\, or insoluble proteins at atomic resolution without a
 n inherent molecular size limitation. MAS NMR experiments to probe protein
  dynamics are extremely versatile and sensitive to motional timescales fro
 m picoseconds to seconds. Over the past decades\, technological advances\,
  developments in experimental design\, and new isotope-labeling approaches
  have further expanded the possibilities of this technique and significant
 ly improved the accuracy of the determined motional parameters.Functionall
 y important sites of proteins often contain aromatic residues. Their side-
 chain motions have therefore long served as valuable indicators of mechani
 stically relevant dynamics in NMR studies. In this thesis\, site-specifica
 lly labeled aromatic residues act as sensitive reporters for MAS NMR studi
 es of protein dynamics. The first part addresses how different environment
 s impact side-chain motion by probing ring flips of phenylalanines and tyr
 osines in crystalline proteins and amyloid fibrils. It provides important 
 insights for the analysis of dynamics obtained in non-native protein envir
 onments and emphasizes the complex factors that determine the timescale of
  internal dynamics. In the second part\, the focus shifts towards methodol
 ogical questions regarding the investigation of protein dynamics by 19F MA
 S NMR. The fluorine nucleus exhibits promising characteristics for NMR stu
 dies but also presents significant challenges\, which is why the full meth
 odological potential of 19F MAS NMR has not been fully realized yet. This 
 work demonstrates that paramagnetic doping can considerably reduce the mea
 surement time and improve the sensitivity of fluorinated samples. Finally\
 , 19F MAS NMR is evaluated as a tool for studying protein side-chain dynam
 ics on the example of tryptophans. The results illustrate the challenges i
 n analyzing such experiments and lay the foundation for further developmen
 t of 19F MAS NMR relaxation studies.Taken together\, this thesis highlight
 s the potential of combining specific isotope labeling\, MAS NMR\, and com
 plementary methods such as crystallography and computational simulations t
 o elucidate internal protein dynamics. The further development of such int
 egrative approaches will be crucial to improving our understanding of comp
 lex mechanisms and protein function.
LOCATION:Central Bldg / O1 / Mondi 2a (I01.O1.008)\, ISTA
ORGANIZER:
SUMMARY:Lea Becker: Thesis Defense: Exploring protein dynamics using specif
 ic labeling approaches for solid-state MAS NMR
URL:https://talks-calendar.ista.ac.at/events/6447
END:VEVENT
BEGIN:VEVENT
DTSTAMP:20260615T191843Z
UID:1782720000@ist.ac.at
DTSTART:20260629T100000
DTEND:20260629T110000
DESCRIPTION:Speaker: Jakub Löwit\nhosted by Krishnendu Chatterjee\nAbstrac
 t: We develop and employ techniques from equivariant algebraic K-theory an
 d related invariants in the context of geometric representation theory\, i
 n both arithmetic and topological situations. We showcase the use of such 
 techniques on the affine Grassmannian Gr\, a space of fundamental interest
  in the geometric Langlands program.
LOCATION:Office Bldg West / Ground floor / Heinzel Seminar Room (I21.EG.101
 ) and Zoom\, ISTA
ORGANIZER:
SUMMARY:Jakub Löwit: Thesis Defense: Equivariant K-theory of affine Grassm
 annians in representation theory and arithmetic
URL:https://talks-calendar.ista.ac.at/events/6500
END:VEVENT
BEGIN:VEVENT
DTSTAMP:20260615T191843Z
UID:1782918000@ist.ac.at
DTSTART:20260701T170000
DTEND:20260701T180000
DESCRIPTION:Speaker: Simone Bombari\nhosted by Matthew Kwan\nAbstract: Arti
 ficial intelligence and machine learning have undergone an unprecedented e
 volution in the past decade\, motivating a research effort toward a theory
  able to capture the qualitative behavior of large-scale neural systems. A
  central puzzle has been the clear benefit of scaling architecture size an
 d overfitting the training set in supervised learning tasks. This evidence
 \, in apparent contradiction with classical statistical learning theory\, 
 pushed researchers to develop a new theory capturing the interplay between
  the algorithmic and architectural bias of training and the specific targe
 t function\, differently from previous methods rooted in uniform stability
 .This approach has enabled a grounded understanding of novel learning regi
 mes\, typically through formal limits where the number of training samples
  $n$\, data dimensions $d$\, and model parameters $p$ grow to infinity at 
 different rates.In this thesis\, we follow this approach\, focusing on the
  trustworthiness of high-dimensional models: properties that are difficult
  to control during training or deployment and often emerge under unpredict
 able or adversarial conditions. In such settings\, it is crucial to formal
 ly ensure a priori the reliability of machine learning systems.First\, we 
 study data memorization\, both as label fitting and as the storage of priv
 ate information about training samples in trained parameters. We prove tha
 t $p = \\Omega(n)$ parameters are sufficient for a deep neural network to 
 memorize a generic set of labels\, and for a model to memorize spurious fe
 atures across training data. We then give evidence that $p = \\Omega(dn)$ 
 parameters are instead necessary for an adversary to reconstruct the full 
 training set from the trained parameters.Second\, we study robustness\, bo
 th to adversarial perturbations and to distribution shift. We first prove 
 that $p = \\Omega(dn)$ parameters can be sufficient for a class of neural 
 networks to overfit the training data while guaranteeing robustness to adv
 ersarial perturbations. Then\, we focus on spurious correlations learning 
 in high-dimensional regression\, studying the effect of the ridge regulari
 zation parameter in the proportional regime $n = \\Theta(d)$\, and connect
 ing it via an equivalence argument to the role of over-parameterization $p
  = \\Omega(n)$ in neural networks. We also investigate the architectural b
 ias of attention-based networks\, showing that they are sensitive to the r
 eplacement of individual words in an embedded sentence\, allowing them to 
 generalize on sentences where the contextual meaning depends on one or few
  words.Finally\, we study differentially private optimization in high-dime
 nsional regimes. We prove that standard private gradient methods do not su
 ffer in the over-parameterized regime $p = \\Omega(n)$\, challenging the c
 urrent wisdom based on stability-derived generalization bounds. We then co
 nsider linear regression in the proportional regime $n = \\Theta(d)$\, sho
 wing that standard private gradient descent can achieve optimal rates unde
 r appropriate hyper-parameter scaling\, such as sufficiently small gradien
 t clipping constants\, whose role is still debated in practice.
LOCATION:Office Bldg West / Ground floor / Heinzel Seminar Room (I21.EG.101
 ) and Zoom\, ISTA
ORGANIZER:
SUMMARY:Simone Bombari: Thesis Defense: Trustworthy Machine Learning in Hig
 h Dimensions
URL:https://talks-calendar.ista.ac.at/events/6499
END:VEVENT
BEGIN:VEVENT
DTSTAMP:20260615T191843Z
UID:1783076400@ist.ac.at
DTSTART:20260703T130000
DTEND:20260703T140000
DESCRIPTION:Speaker: Xin Tong\nhosted by Carl Goodrich\nAbstract: Embryo cl
 eavage — a series of rapid\, reductive cell divisions — is the first m
 orphogenetic movement following fertilisation. It lays the foundation for 
 subsequent developmental events\, including gastrulation\, germ layer spec
 ification\, organogenesis\, and the establishment of the overall body plan
 . Two major modes of cleavage exist in the animal kingdom: holoblastic (co
 mplete) and meroblastic (incomplete) cleavage. Because holoblastic cleavag
 e resembles canonical cytokinesis\, its biochemical and mechanical basis h
 as been extensively studied — an annular contractile ring forms at the e
 quator and constricts in a purse-string-like manner\, leading to the compl
 ete separation of two daughter cells. In contrast\, although meroblastic c
 leavage occurs widely across the animal kingdom (e.g. in fish\, reptiles\,
  birds\, and cephalopod molluscs)\, its mechanical basis remains largely u
 nclear. During meroblastic cleavage\, the cytokinetic furrow forms only at
  one pole and does not traverse the entire embryo\, raising the question o
 f how cytokinesis proceeds in the absence of a closed contractile ring. Mo
 reover\, the resulting daughter cells are not fully separated from the und
 erlying yolk compartment\, and how these blastomeres are subsequently cell
 ularised remains unknown. This thesis takes the zebrafish as a model organ
 ism to address both questions. The first part characterises the biochemica
 l and mechanical mechanisms underlying non-canonical meroblastic cleavage\
 , revealing a two-phase process in which actomyosin cable contraction and 
 cadherin-mediated membrane adhesion act sequentially to drive furrow ingre
 ssion and invagination. The second part sheds light on the spatiotemporal 
 dynamics by which individual blastomeres become cellularised\, and uncover
 s a previously unrecognised contribution of central blastomeres to the yol
 k syncytial layer.
LOCATION:Central Bldg / O1 / Mondi 3 (I01.O1.010) \, ISTA
ORGANIZER:
SUMMARY:Xin Tong: Thesis Defense: Towards a deeper understanding of merobla
 stic cleavage - biochemical mechanisms of partial cytokinesis and cellular
 ization in zebrafish embryogenesis
URL:https://talks-calendar.ista.ac.at/events/6519
END:VEVENT
BEGIN:VEVENT
DTSTAMP:20260615T191843Z
UID:1783598400@ist.ac.at
DTSTART:20260709T140000
DTEND:20260709T150000
DESCRIPTION:Speaker: David Michalik\nhosted by Eva Benková\nAbstract: MDA5
  is an antiviral protein that is activated by long dsRNA molecules. This R
 NA can be of various origins such as viral\, bacterial\, synthetic or\, up
 on certain stimuli\, even cellularsources. Upon sensing immunogenic RNA\, 
 MDA5 coats the dsRNA\, forming a filament. Filament formation brings CARD 
 domains into close proximity\, resulting in CARD oligomerisation. These ol
 igomers are not involved in the RNA recognition but are crucial for activa
 tion of downstream signalling. MDA5 activation leads to a type I interfero
 n response and apoptosis. Loss of function mutations are linked to recurre
 nt\, life-threatening viral infections whereas gain of function mutations 
 are linked to monogenic and polygenic autoimmune diseases. For these reaso
 ns\, MDA5 needs to be tightly regulated by various mechanisms including po
 st-translational modifications\, post-transcriptional modifications\, and 
 protein-protein interactions. In the first chapter\, we focused on the cha
 racterisation of novel phosphorylation sites in MDA5. Using biochemistry\,
  molecular dynamics simulations\, virology\, cell biology and mass spectro
 metry\, we characterised two phosphorylation sites that altered MDA5 activ
 ity andwere regulated in cells upon EMCV infection or IFN-β treatment. Fu
 rthermore\, we discovered an additional phosphorylation site in a non-cons
 erved region\, which is upregulated under both stimuli.In the second chapt
 er\, we focused on biochemical\, biophysical\, and structural characterisa
 tion of various RNAs and MDA5 filaments assembled on them. We discovered t
 hat poly I:C\, a synthetic RNA used to activate an immune response in cell
 s\, has distinct physicochemical characteristics and is not a true mimic o
 f viral RNA\, as is often described. Additionally\, we used cryogenic elec
 tron microscopy to gain structural insights into which RNA features define
  a strong activator of MDA5. To address this\, we compared poly I:C to pol
 y A:U\, an RNA that does not activate an MDA5 signalling in cells\, and to
  a virus-derived dsRNA. In the last chapter\, we focused on activators of 
 MDA5 signalling. In the first part\, we present our efforts to isolate and
  identify a small molecule agonist of MDA5. In the last part\, we describ
 ed our contribution to a collaborative project during which we characteris
 ed the interaction of ANXA2 with MDA5 and activation of MDA5 by a small mo
 lecule.
LOCATION:Sunstone Bldg / Ground floor / Big Seminar Room B / 63 seats (I23.
 EG.102)\, ISTA
ORGANIZER:
SUMMARY:David Michalik: Thesis Defense: Mechanistic insights into MDA5 sele
 ctivity and regulation
URL:https://talks-calendar.ista.ac.at/events/6508
END:VEVENT
END:VCALENDAR