Universitätspublikationen
Refine
Year of publication
- 2023 (175) (remove)
Document Type
- Preprint (117)
- Article (51)
- Contribution to a Periodical (2)
- Doctoral Thesis (2)
- Bachelor Thesis (1)
- Conference Proceeding (1)
- Master's Thesis (1)
Has Fulltext
- yes (175)
Is part of the Bibliography
- no (175)
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)
Institute
- Informatik (175) (remove)
Large language models have become widely available to the general public, especially due to ChatGPT's release. Consequently, the AI community has invested much effort into recreating language models of the same caliber as ChatGPT, since the latter is still a technical blackbox. This thesis aims to contribute to that cause by proposing R.O.B.E.R.T., a Robotic Operating Buddy for Efficiency, Research and Teaching. In doing so, it presents a first implementation of a lightweight environment which produces tailor-made, instruction-following language models with a heavy focus on conversational capabilities that instruct themselves into a given domain-context. Within this environment, the generation of datasets, the fine-tuning process and finally the inference of a unique R.O.B.E.R.T. instance are all carried out as part of an automated pipeline.
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.
The fundamental structure of cortical networks arises early in development prior to the onset of sensory experience. However, how endogenously generated networks respond to the onset of sensory experience, and how they form mature sensory representations with experience remains unclear. Here we examine this "nature-nurture transform" using in vivo calcium imaging in ferret visual cortex. At eye-opening, visual stimulation evokes robust patterns of cortical activity that are highly variable within and across trials, severely limiting stimulus discriminability. Initial evoked responses are distinct from spontaneous activity of the endogenous network. Visual experience drives the development of low-dimensional, reliable representations aligned with spontaneous activity. A computational model shows that alignment of novel visual inputs and recurrent cortical networks can account for the emergence of reliable visual representations.
An excess of J/ψ yield at very low transverse momentum (pT<0.3 GeV/c), originating from coherent photoproduction, is observed in peripheral and semicentral hadronic Pb−Pb collisions at a center-of-mass energy per nucleon pair of sNN−−−√=5.02 TeV. The measurement is performed with the ALICE detector via the dimuon decay channel at forward rapidity (2.5<y<4). The nuclear modification factor at very low pT and the coherent photoproduction cross section are measured as a function of centrality down to the 10% most central collisions. These results extend the previous study at sNN−−−√=2.76 TeV, confirming the clear excess over hadronic production in the pT range 0−0.3 GeV/c and the centrality range 70−90%, and establishing an excess with a significance greater than 5σ also in the 50−70% and 30−50% centrality ranges. The results are compared with earlier measurements at sNN−−−√=2.76 TeV and with different theoretical predictions aiming at describing how coherent photoproduction occurs in hadronic interactions with nuclear overlap.
The fundamental structure of cortical networks arises early in development prior to the onset of sensory experience. However, how endogenously generated networks respond to the onset of sensory experience, and how they form mature sensory representations with experience remains unclear. Here we examine this ‘nature-nurture transform’ using in vivo calcium imaging in ferret visual cortex. At eye-opening, visual stimulation evokes robust patterns of cortical activity that are highly variable within and across trials, severely limiting stimulus discriminability. Initial evoked responses are distinct from spontaneous activity of the endogenous network. Visual experience drives the development of low-dimensional, reliable representations aligned with spontaneous activity. A computational model shows that alignment of novel visual inputs and recurrent cortical networks can account for the emergence of reliable visual representations.
An excess of J/ψ yield at very low transverse momentum (pT<0.3 GeV/c), originating from coherent photoproduction, is observed in peripheral and semicentral hadronic Pb–Pb collisions at a center-of-mass energy per nucleon pair of sNN=5.02 TeV. The measurement is performed with the ALICE detector via the dimuon decay channel at forward rapidity (2.5<y<4). The nuclear modification factor at very low pT and the coherent photoproduction cross section are measured as a function of centrality down to the 10% most central collisions. These results extend the previous study at sNN=2.76 TeV, confirming the clear excess over hadronic production in the pT range 0−0.3 GeV/c and the centrality range 70–90%, and establishing an excess with a significance greater than 5σ also in the 50–70% and 30–50% centrality ranges. The results are compared with earlier measurements at sNN=2.76 TeV and with different theoretical predictions aiming at describing how coherent photoproduction occurs in hadronic interactions with nuclear overlap.
Non-coding variations located within regulatory elements may alter gene expression by modifying Transcription Factor (TF) binding sites and thereby lead to functional consequences like various traits or diseases. To understand these molecular mechanisms, different TF models are being used to assess the effect of DNA sequence variations, such as Single Nucleotide Polymorphisms (SNPs). However, few statistical approaches exist to compute statistical significance of results but they often are slow for large sets of SNPs, such as data obtained from a genome-wide association study (GWAS) or allele-specific analysis of chromatin data.
Results We investigate the distribution of maximal differential TF binding scores for general computational models that assess TF binding. We find that a modified Laplace distribution can adequately approximate the empirical distributions. A benchmark on in vitro and in vivo data sets showed that our new approach improves on an existing method in terms of performance and speed. In applications on large sets of eQTL and GWAS SNPs we could illustrate the usefulness of the novel statistic to highlight cell type specific regulators and TF target genes.
Conclusions Our approach allows the evaluation of DNA changes that induce differential TF binding in a fast and accurate manner, permitting computations on large mutation data sets. An implementation of the novel approach is freely available at https://github.com/SchulzLab/SNEEP.
W±-boson production in p–Pb collisions at √sNN = 8.16 TeV and Pb–Pb collisions at √sNN = 5.02 TeV
(2023)
The production of the W± bosons measured in p–Pb collisions at a centreof-mass energy per nucleon–nucleon collision √sNN = 8.16 TeV and Pb–Pb collisions at √sNN = 5.02 TeV with ALICE at the LHC is presented. The W± bosons are measured via their muonic decay channel, with the muon reconstructed in the pseudorapidity region −4 < ηµ lab < −2.5 with transverse momentum p µ T > 10 GeV/c. While in Pb–Pb collisions the measurements are performed in the forward (2.5 < yµ cms < 4) rapidity region, in p–Pb collisions, where the centre-of-mass frame is boosted with respect to the laboratory frame, the measurements are performed in the backward (−4.46 < yµ cms < −2.96) and forward (2.03 < yµ cms < 3.53) rapidity regions. The W− and W+ production cross sections, leptoncharge asymmetry, and nuclear modification factors are evaluated as a function of the muon rapidity. In order to study the production as a function of the p–Pb collision centrality, the production cross sections of the W− and W+ bosons are combined and normalised to the average number of binary nucleon–nucleon collision hNcolli. In Pb–Pb collisions, the same measurements are presented as a function of the collision centrality. Study of the binary scaling of the W±-boson cross sections in p–Pb and Pb–Pb collisions is also reported. The results are compared with perturbative QCD calculations, with and without nuclear modifications of the Parton Distribution Functions (PDFs), as well as with available data at the LHC. Significant deviations from the theory expectations are found in the two collision systems, indicating that the measurements can provide additional constraints for the determination of nuclear PDFs and in particular of the light-quark distributions.