Universitätspublikationen
Refine
Year of publication
- 2023 (182) (remove)
Document Type
- Preprint (118)
- Article (56)
- Bachelor Thesis (2)
- Contribution to a Periodical (2)
- Doctoral Thesis (2)
- Conference Proceeding (1)
- Master's Thesis (1)
Has Fulltext
- yes (182)
Is part of the Bibliography
- no (182)
Keywords
- Hadron-Hadron Scattering (4)
- Quark-Gluon Plasma (4)
- Jets and Jet Substructure (3)
- Collective Flow (2)
- Heavy Ion Experiments (2)
- Active learning (1)
- Boosted Jets (1)
- Chatbot (1)
- Continual deep learning (1)
- Dialog Generation (1)
- Hadron-Hadron Scattering Heavy (1)
- Heavy Ion Experiment (1)
- Heavy Quark Production (1)
- Jet Physics (1)
- Jet Substructure (1)
- Jets (1)
- Large Language Models (1)
- Lifelong machine learning (1)
- MIMIC-III (1)
- Material budget (1)
- Open set recognition (1)
- Open world learning (1)
- Particle Correlations and Fluctuations (1)
- Particle and Resonance Production (1)
- Properties of Hadrons (1)
- Prophet Inequalities (1)
- QCD (1)
- Quark Deconfinement (1)
- Quark Production (1)
- Quarkonium (1)
- Self-Instruct (1)
- Stochastic Probing (1)
- Testing (1)
- Vector Boson Production (1)
- data analysis (1)
- detector (1)
- experimental results (1)
- machine learning (1)
- prediction quality (1)
- stroke (1)
Institute
- Informatik (182) (remove)
Assessing communicative accommodation in the context of large language models : a semiotic approach
(2023)
Recently, significant strides have been made in the ability of transformer-based chatbots to hold natural conversations. However, despite a growing societal and scientific relevancy, there are few frameworks systematically deriving what it means for a chatbot conversation to be natural. The present work approaches this question through the phenomenon of communicative accommodation/interactive alignment. While there is existing research suggesting that humans adapt communicatively to technologies, the aim of this work is to explore the accommodation of AI-chatbots to an interlocutor. Its research interest is twofold: Firstly, the structural ability of the transformer-architecture to support accommodative behavior is assessed using a frame constructed in accordance with existing accommodationtheories.
This results in hypotheses to be tested empirically. Secondly, since effective accommodation produces the same outcomes, regardless of technical implementation, a behavioral experiment is proposed. Existing quantifications of accommodation are reconciled,
extended, and modified to apply them to nonhuman-interlocutors. Thus, a measurement scheme is suggested which evaluates textual data from text-only, double-blind interactions between chatbots and humans, chatbots and chatbots and humans and humans. Using the generated human-to-human convergence data as a reference, the degree of artificial accommodation can be evaluated. Accommodation as a central facet of artificial interactivity can thus be evaluated directly against its theoretical paradigm, i.e. human interaction. In case that subsequent examinations show that chatbots effectively do not accommodate, there may be a new form of algorithmic bias, emerging from the aggregate accommodation towards chatbots but not towards humans. Thus, existing, hegemonic semantics could be cemented through chatbot-learning. Meanwhile, the ability to effectively accommodate would render chatbots vastly more susceptible to misuse.
The free energy of TAP-solutions for the SK-model of mean field spin glasses can be expressed as a nonlinear functional of local terms: we exploit this feature in order to contrive abstract REM-like models which we then solve by a classical large deviations treatment. This allows to identify the origin of the physically unsettling quadratic (in the inverse of temperature) correction to the Parisi free energy for the SK-model, and formalizes the true cavity dynamics which acts on TAP-space, i.e. on the space of TAP-solutions. From a non-spin glass point of view, this work is the first in a series of refinements which addresses the stability of hierarchical structures in models of evolving populations.
The production of K∗(892)± meson resonance is measured at midrapidity (|y|<0.5) in Pb-Pb collisions at sNN−−−√=5.02 TeV using the ALICE detector at the LHC. The resonance is reconstructed via its hadronic decay channel K∗(892)±→K0Sπ±. The transverse momentum distributions are obtained for various centrality intervals in the pT range of 0.4-16 GeV/c. The reported measurements of integrated yields, mean transverse momenta, and particle yield ratios are consistent with previous ALICE measurements for K∗(892)0. The pT-integrated yield ratio 2K∗(892)±/(K++K−) in central Pb-Pb collisions shows a significant suppression (9.3σ) relative to pp collisions. Thermal model calculations overpredict the particle yield ratio. Although both simulations consider the hadronic phase, only HRG-PCE accurately represents the measurements, whereas MUSIC+SMASH tends to overpredict them. These observations, along with the kinetic freeze-out temperatures extracted from the yields of light-flavored hadrons using the HRG-PCE model, indicate a finite hadronic phase lifetime, which increases towards central collisions. The pT-differential yield ratios 2K∗(892)±/(K++K−) and 2K∗(892)±/(π++π−) are suppressed by up to a factor of five at pT<2 GeV/c in central Pb-Pb collisions compared to pp collisions at s√= 5.02 TeV. Both particle ratios and are qualitatively consistent with expectations for rescattering effects in the hadronic phase. The nuclear modification factor shows a smooth evolution with centrality and is below unity at pT>8 GeV/c, consistent with measurements for other light-flavored hadrons. The smallest values are observed in most central collisions, indicating larger energy loss of partons traversing the dense medium.
A new, more precise measurement of the Λ hyperon lifetime is performed using a large data sample of Pb–Pb collisions at √sNN p ¼ 5.02 TeV with ALICE. The Λ and Λ¯ hyperons are reconstructed at midrapidity using their two-body weak decay channel Λ → p þ π− and Λ¯ → p¯ þ πþ. The measured value of the Λ lifetime is τΛ ¼ ½261.07 0.37ðstat:Þ 0.72ðsyst:Þ ps. The relative difference between the lifetime of Λ and Λ¯ , which represents an important test of CPT invariance in the strangeness sector, is also measured. The obtained value ðτΛ − τΛ¯Þ=τΛ ¼ 0.0013 0.0028ðstat:Þ 0.0021ðsyst:Þ is consistent with zero within the uncertainties. Both measurements of the Λ hyperon lifetime and of the relative difference between τΛ and τΛ¯ are in agreement with the corresponding world averages of the Particle Data Group and about a factor of three more precise.
First measurement of Λ+c production down to pT = 0 in pp and p-Pb collisions at √𝑠NN = 5.02 TeV
(2023)
The production of prompt +c baryons has been measured at midrapidity in the transverse momentum interval 0 < pT < 1 GeV/c for the first time, in pp and p–Pb collisions at a center-of-mass energy per nucleon-nucleon collision √sNN = 5.02 TeV. The measurement was performed in the decay channel +c → pK0S by applying new decay reconstruction techniques using a Kalman-Filter vertexing algorithm and adopting a machine-learning approach for the candidate selection. The pT -integrated +c production cross sections in both collision systems were determined and used along with the measured yields in Pb–Pb collisions to compute the pT -integrated nuclear modification factors RpPb and RAA of +c baryons, which are compared to model calculations that consider nuclear modification of the parton distribution functions. The +c /D0 baryon-to-meson yield ratio is reported for pp and p–Pb collisions. Comparisons with models that include modified hadronization processes are presented, and the implications of the results on the understanding of charm hadronization in hadronic collisions are discussed. A significant (3.7σ) modification of the mean transverse momentum of + c baryons is seen in p–Pb collisions with respect to pp collisions, while the pT -integrated +c /D0 yield ratio was found to be consistent between the two collision systems within the uncertainties.
The measurement of the production of deuterons, tritons and 3He and their antiparticles in Pb-Pb collisions at √sNN = 5.02 TeV is presented in this article. The measurements are carried out at midrapidity (y|< 0.5) as a function of collision centrality using the ALICE detector. The pT-integrated yields, the coalescence parameters and the ratios to protons and antiprotons are reported and compared with nucleosynthesis models. The comparison of these results in different collision systems at different center-of-mass collision energies reveals a suppression of nucleus production in small systems. In the Statistical Hadronisation Model framework, this can be explained by a small correlation volume where the baryon number is conserved, as already shown in previous fluctuation analyses. However, a different size of the correlation volume is required to describe the proton yields in the same data sets. The coalescence model can describe this suppression by the fact that the wave functions of the nuclei are large and the fireball size starts to become comparable and even much smaller than the actual nucleus at low multiplicities.
The knowledge of the material budget with a high precision is fundamental for measurements of direct photon production using the photon conversion method due to its direct impact on the total systematic uncertainty. Moreover, it influences many aspects of the charged-particle reconstruction performance. In this article, two procedures to determine data-driven corrections to the material-budget description in ALICE simulation software are developed. One is based on the precise knowledge of the gas composition in the Time Projection Chamber. The other is based on the robustness of the ratio between the produced number of photons and charged particles, to a large extent due to the approximate isospin symmetry in the number of produced neutral and charged pions. Both methods are applied to ALICE data allowing for a reduction of the overall material budget systematic uncertainty from 4.5% down to 2.5%. Using these methods, a locally correct material budget is also achieved. The two proposed methods are generic and can be applied to any experiment in a similar fashion.
Analysis of machine learning prediction quality for automated subgroups within the MIMIC III dataset
(2023)
The motivation for this master’s thesis is to explore the potential of predictive data analytics in the field of medicine. For this, the MIMIC-III dataset offers an extensive foundation for the construction of prediction models, including Random Forest, XGBOOST, and deep learning networks. These models were implemented to forecast the mortality of 2,655 stroke patients.
The first part of the thesis involved conducting a comprehensive data analysis of the filtered MIMIC-III dataset.
Subsequently, the effectiveness and fairness of the predictive models were evaluated. Although the performance levels of the developed models did not match those reported in related research, their potential became evident. The results obtained demonstrated promising capabilities and highlighted the effectiveness of the applied methodologies. Moreover, the feature relevance within the XGBOOST model was examined to increase model explainability.
Finally, relevant subgroups were identified to perform a comparative analysis of the prediction performance across these subgroups. While this approach can be regarded as a valuable methodology, it was not possible to investigate underlying reasons for potential unfairness across clusters. Inside the test data, not enough instances remained per subgroup for further fairness or feature relevance analysis.
In conclusion, the implementation of an alternative use case with a higher patient count is recommended.
The code for this analysis is made available via a GitHub repository and includes a frontend to visualize the results.
We study threshold testing, an elementary probing model with the goal to choose a large value out of n i.i.d. random variables. An algorithm can test each variable X_i once for some threshold t_i, and the test returns binary feedback whether X_i ≥ t_i or not. Thresholds can be chosen adaptively or non-adaptively by the algorithm. Given the results for the tests of each variable, we then select the variable with highest conditional expectation. We compare the expected value obtained by the testing algorithm with expected maximum of the variables. Threshold testing is a semi-online variant of the gambler’s problem and prophet inequalities. Indeed, the optimal performance of non-adaptive algorithms for threshold testing is governed by the standard i.i.d. prophet inequality of approximately 0.745 + o(1) as n → ∞. We show how adaptive algorithms can significantly improve upon this ratio. Our adaptive testing strategy guarantees a competitive ratio of at least 0.869 - o(1). Moreover, we show that there are distributions that admit only a constant ratio c < 1, even when n → ∞. Finally, when each box can be tested multiple times (with n tests in total), we design an algorithm that achieves a ratio of 1 - o(1).
Current deep learning methods are regarded as favorable if they empirically perform well on dedicated test sets. This mentality is seamlessly reflected in the resurfacing area of continual learning, where consecutively arriving data is investigated. The core challenge is framed as protecting previously acquired representations from being catastrophically forgotten. However, comparison of individual methods is nevertheless performed in isolation from the real world by monitoring accumulated benchmark test set performance. The closed world assumption remains predominant, i.e. models are evaluated on data that is guaranteed to originate from the same distribution as used for training. This poses a massive challenge as neural networks are well known to provide overconfident false predictions on unknown and corrupted instances. In this work we critically survey the literature and argue that notable lessons from open set recognition, identifying unknown examples outside of the observed set, and the adjacent field of active learning, querying data to maximize the expected performance gain, are frequently overlooked in the deep learning era. Hence, we propose a consolidated view to bridge continual learning, active learning and open set recognition in deep neural networks. Finally, the established synergies are supported empirically, showing joint improvement in alleviating catastrophic forgetting, querying data, selecting task orders, while exhibiting robust open world application.