@phdthesis{68,
abstract = {The most common assumption made in statistical learning theory is the assumption of the independent and identically distributed (i.i.d.) data. While being very convenient mathematically, it is often very clearly violated in practice. This disparity between the machine learning theory and applications underlies a growing demand in the development of algorithms that learn from dependent data and theory that can provide generalization guarantees similar to the independent situations. This thesis is dedicated to two variants of dependencies that can arise in practice. One is a dependence on the level of samples in a single learning task. Another dependency type arises in the multi-task setting when the tasks are dependent on each other even though the data for them can be i.i.d. In both cases we model the data (samples or tasks) as stochastic processes and introduce new algorithms for both settings that take into account and exploit the resulting dependencies. We prove the theoretical guarantees on the performance of the introduced algorithms under different evaluation criteria and, in addition, we compliment the theoretical study by the empirical one, where we evaluate some of the algorithms on two real world datasets to highlight their practical applicability.},
author = {Zimin, Alexander},
pages = {92},
publisher = {IST Austria},
title = {{Learning from dependent data}},
doi = {10.15479/AT:ISTA:TH1048},
year = {2018},
}
@phdthesis{69,
abstract = {A qubit, a unit of quantum information, is essentially any quantum mechanical two-level system which can be coherently controlled. Still, to be used for computation, it has to fulfill criteria. Qubits, regardless of the system in which they are realized, suffer from decoherence. This leads to loss of the information stored in the qubit. The upper bound of the time scale on which decoherence happens is set by the spin relaxation time. In this thesis I studied a two-level system consisting of a Zeeman-split hole spin confined in a quantum dot formed in a Ge hut wire. Such Ge hut wires have emerged as a promising material system for the realization of spin qubits, due to the combination of two significant properties: long spin coherence time as expected for group IV semiconductors due to the low hyperfine interaction and a strong valence band spin-orbit coupling. Here, I present how to fabricate quantum dot devices suitable for electrical transport measurements. Coupled quantum dot devices allowed the realization of a charge sensor, which is electrostatically and tunnel coupled to a quantum dot. By integrating the charge sensor into a radio-frequency reflectometry setup, I performed for the first time single-shot readout measurements of hole spins and extracted the hole spin relaxation times in Ge hut wires.},
author = {Vukušić, Lada},
pages = {103},
publisher = {IST Austria},
title = {{Charge sensing and spin relaxation times of holes in Ge hut wires}},
doi = {10.15479/AT:ISTA:TH_1047},
year = {2018},
}
@phdthesis{197,
abstract = {Modern computer vision systems heavily rely on statistical machine learning models, which typically require large amounts of labeled data to be learned reliably. Moreover, very recently computer vision research widely adopted techniques for representation learning, which further increase the demand for labeled data. However, for many important practical problems there is relatively small amount of labeled data available, so it is problematic to leverage full potential of the representation learning methods. One way to overcome this obstacle is to invest substantial resources into producing large labelled datasets. Unfortunately, this can be prohibitively expensive in practice. In this thesis we focus on the alternative way of tackling the aforementioned issue. We concentrate on methods, which make use of weakly-labeled or even unlabeled data. Specifically, the first half of the thesis is dedicated to the semantic image segmentation task. We develop a technique, which achieves competitive segmentation performance and only requires annotations in a form of global image-level labels instead of dense segmentation masks. Subsequently, we present a new methodology, which further improves segmentation performance by leveraging tiny additional feedback from a human annotator. By using our methods practitioners can greatly reduce the amount of data annotation effort, which is required to learn modern image segmentation models. In the second half of the thesis we focus on methods for learning from unlabeled visual data. We study a family of autoregressive models for modeling structure of natural images and discuss potential applications of these models. Moreover, we conduct in-depth study of one of these applications, where we develop the state-of-the-art model for the probabilistic image colorization task.},
author = {Kolesnikov, Alexander},
pages = {113},
publisher = {IST Austria},
title = {{Weakly-Supervised Segmentation and Unsupervised Modeling of Natural Images}},
doi = {10.15479/AT:ISTA:th_1021},
year = {2018},
}
@phdthesis{83,
abstract = {A proof system is a protocol between a prover and a verifier over a common input in which an honest prover convinces the verifier of the validity of true statements. Motivated by the success of decentralized cryptocurrencies, exemplified by Bitcoin, the focus of this thesis will be on proof systems which found applications in some sustainable alternatives to Bitcoin, such as the Spacemint and Chia cryptocurrencies. In particular, we focus on proofs of space and proofs of sequential work.
Proofs of space (PoSpace) were suggested as more ecological, economical, and egalitarian alternative to the energy-wasteful proof-of-work mining of Bitcoin. However, the state-of-the-art constructions of PoSpace are based on sophisticated graph pebbling lower bounds, and are therefore complex. Moreover, when these PoSpace are used in cryptocurrencies like Spacemint, miners can only start mining after ensuring that a commitment to their space is already added in a special transaction to the blockchain. Proofs of sequential work (PoSW) are proof systems in which a prover, upon receiving a statement x and a time parameter T, computes a proof which convinces the verifier that T time units had passed since x was received. Whereas Spacemint assumes synchrony to retain some interesting Bitcoin dynamics, Chia requires PoSW with unique proofs, i.e., PoSW in which it is hard to come up with more than one accepting proof for any true statement. In this thesis we construct simple and practically-efficient PoSpace and PoSW. When using our PoSpace in cryptocurrencies, miners can start mining on the fly, like in Bitcoin, and unlike current constructions of PoSW, which either achieve efficient verification of sequential work, or faster-than-recomputing verification of correctness of proofs, but not both at the same time, ours achieve the best of these two worlds.},
author = {Abusalah, Hamza M},
pages = {59},
publisher = {IST Austria},
title = {{Proof systems for sustainable decentralized cryptocurrencies}},
doi = {10.15479/AT:ISTA:TH_1046},
year = {2018},
}
@phdthesis{9,
abstract = {Immune cells migrating to the sites of infection navigate through diverse tissue architectures and switch their migratory mechanisms upon demand. However, little is known about systemic regulators that could allow the acquisition of these mechanisms. We performed a genetic screen in Drosophila melanogaster to identify regulators of germband invasion by embryonic macrophages into the confined space between the ectoderm and mesoderm. We have found that bZIP circadian transcription factors (TFs) Kayak (dFos) and Vrille (dNFIL3) have opposite effects on macrophage germband infiltration: Kayak facilitated and Vrille inhibited it. These TFs are enriched in the macrophages during migration and genetically interact to control it. Kayak sets a less coordinated mode of migration of the macrophage group and increases the probability and length of Levy walks. Intriguingly, the motility of kayak mutant macrophages was also strongly affected during initial germband invasion but not along another less confined route. Inhibiting Rho1 signaling within the tail ectoderm partially rescued the Kayak mutant phenotype, strongly suggesting that migrating macrophages have to overcome a barrier imposed by the stiffness of the ectoderm. Also, Kayak appeared to be important for the maintenance of the round cell shape and the rear edge translocation of the macrophages invading the germband. Complementary to this, the cortical actin cytoskeleton of Kayak- deficient macrophages was strongly affected. RNA sequencing revealed the filamin Cheerio and tetraspanin TM4SF to be downstream of Kayak. Chromatin immunoprecipitation and immunostaining revealed that the formin Diaphanous is another downstream target of Kayak. Immunostaining revealed that the formin Diaphanous is another downstream target of Kayak. Indeed, Cheerio, TM4SF and Diaphanous are required within macrophages for germband invasion, and expression of constitutively active Diaphanous in macrophages was able to rescue the kayak mutant phenotype. Moreover, Cher and Diaphanous are also reduced in the macrophages overexpressing Vrille. We hypothesize that Kayak, through its targets, increases actin polymerization and cortical tension in macrophages and thus allows extra force generation necessary for macrophage dissemination and migration through confined stiff tissues, while Vrille counterbalances it.},
author = {Belyaeva, Vera},
pages = {96},
publisher = {IST Austria},
title = {{Transcriptional regulation of macrophage migration in the Drosophila melanogaster embryo }},
doi = {10.15479/AT:ISTA:th1064},
year = {2018},
}