• Digital Humanities and Cultural Heritage
• Publications close
• Other research products close
• CH close
• ID close
• English close

Whether a document is of historical or contemporary significance, typography plays a crucial role in its composition. From the early days of modern printing, typographic techniques have evolved and transformed, resulting in changes to the features of typography. By analyzing these features, we can gain insights into specific time periods, geographical locations, and messages conveyed through typography. Therefore, in this paper, we aim to investigate the feasibility of training a model to classify serif typeswithout knowledge of the font and character. We also investigate how to train a vectorial-based image model able to group together fonts with similar features. Specifically, we compare the use of state-of-theart image classification methods, such as the EfficientNet-B2 and the Vision Transformer Base model with different patch sizes, and the state-of-the-art fine-grained image classification method, TransFG, on the serif classification task. We also evaluate the use of the DeepSVG model to learn to group fonts with similar features. Our investigation reveals that fine-grained image classification methods are better suited for the serif classification tasks and that leveraging the character labels helps to learn more meaningful font similarities. This repository contains: - Paper published in the Journal of data mining and digital humanities:WasimEtAl_Toward_Automatic_Typography_Analysis__Serif_Classification_and_Font_Similarities.pdf - Two datasets: The first for serif classification consisting of 126666 training and 2914 font-independent testing images in raster format. The second dataset for svg based similarity learning consists of 124010 training and 2914 font-independent testing images. The images have been categorized into sans-serif, linear-serif, slab-serif, and triangular serif by a designer at the EPFL+ECAL Lab font_serif_dataset.zipfont_svg_dataset.zip - model weights:serif_B_16_long_checkpoint.bin: refers to TransFG ViT B/16 Backendserif_B_32_long_checkpoint.bin: refers to TransFG ViT B/32 Backend ViT-B_16.npz ViT-B_32.npz are publicly available at: https://github.com/TACJu/TransFGand were used for training TransFG ViT B/16 Backend and TransFG ViT B/32 Backend - code:https://github.com/TalalWasim/GEST-Serif

Drawing on fieldwork findings, this paper presents the transformation of Sulaiman-Too Mountain in Osh (Kyrgyzstan), which has been going through a process of culturalisation since the mid-20th century. My aim is to show the museification and heritagisation processes, which has been taking place during the Soviet era and more recently. Accordingly, the site emerges as a palimpsest whose meaning changes with political contexts, combining local and global dynamics.

As one of the most effective methods to improve the accuracy and robustness of speech tasks, the audio–visual fusion approach has recently been introduced into the field of Keyword Spotting (KWS). However, existing audio–visual keyword spotting models are limited to detecting isolated words, while keyword spotting for unconstrained speech is still a challenging problem. To this end, an Audio–Visual Keyword Transformer (AVKT) network is proposed to spot keywords in unconstrained video clips. The authors present a transformer classifier with learnable CLS tokens to extract distinctive keyword features from the variable‐length audio and visual inputs. The outputs of audio and visual branches are combined in a decision fusion module. As humans can easily notice whether a keyword appears in a sentence or not, our AVKT network can detect whether a video clip with a spoken sentence contains a pre‐specified keyword. Moreover, the position of the keyword is localised in the attention map without additional position labels. Exper- imental results on the LRS2‐KWS dataset and our newly collected PKU‐KWS dataset show that the accuracy of AVKT exceeded 99% in clean scenes and 85% in extremely noisy conditions. The code is available at https://github.com/jialeren/AVKT. CAAI Transactions on Intelligence Technology, 9 (1) ISSN:2468-2322

Retrieval-augmented generation (RAG) can enhance the generation quality of large language models (LLMs) by incorporating external token databases. However, retrievals from large databases can constitute a substantial portion of the overall generation time, particularly when retrievals are periodically performed to align the retrieved content with the latest states of generation. In this paper, we introduce PipeRAG, a novel algorithm-system co-design approach to reduce generation latency and enhance generation quality. PipeRAG integrates (1) pipeline parallelism to enable concurrent retrieval and generation processes, (2) flexible retrieval intervals to maximize the efficiency of pipeline parallelism, and (3) a performance model to automatically balance retrieval quality and latency based on the generation states and underlying hardware. Our evaluation shows that, by combining the three aforementioned methods, PipeRAG achieves up to 2.6$\times$ speedup in end-to-end generation latency while improving generation quality. These promising results showcase the effectiveness of co-designing algorithms with underlying systems, paving the way for the adoption of PipeRAG in future RAG systems.

Abstracts of all the sessions of a monthly seminar dedicated to the governance of digital infrastructures that took place between September 2021 and May 2022 at the Laboratory for the history of science and technology (LHST) of the École polytechnique fédérale de Lausanne (EPFL, Switzerland) while Pierre Mounier (EHESS, OpenEdition) was visiting scholar.

The site of La Roche-à-Pierrot in Saint-Césaire (Charente-Maritime, France) produced a succession of Mousterian, Châtelperronian and Aurignacian occupations, and continues to play a central role in debates concerning the Middle-to-Upper Palaeolithic transition. The source of controversy surrounding the site relates to ambiguities concerning the overall archaeological sequence, the cultural association of the human remains found at the site and the limited number of robust absolute dates. Here, we present the results of a multiscalar, multiproxy geoarchaeological investigation of the site's sedimentary sequence. Our study integrates geomorphology, field lithostratigraphy, microstratigraphy, geochemistry and absolute dating methods (radiocarbon and optically stimulated luminescence) designed to characterize site formation processes. We propose a site formation model involving the evolution of a karstified limestone cliff from a semi-closed system to an exposed slope deposit, with sediments at the base of the cliff accumulating under periglacial conditions of MIS3, broadly between ca. 59.9 ± 3.9 ka and ca. 37.7 ka. The lowermost Mousterian occupations took place in a semi-closed, sheltered space in which sedimentary rates were low, involving mainly cryoclastic roof spall and episodic percolation of fine-grained sediment. This depositional environment generated a Mousterian archaeological palimpsest in the proximal area and translocated downslope. Subsequent Mousterian, Châtelperronian and Aurignacian remains embedded within diamictons were deposited in an open-air context, on a sloping surface at the apex of a palaeotalus, and possibly above the site, at the top of the cliff. The resulting diachronous surfaces and immature deposits were subject to solifluction and slopewash, with low sedimentary rates. Our geoarchaeological study sheds new light on previously noted spatial inconsistencies in the archaeological sequence and constitutes a rigorous framework for further archaeological research at the site, highlighting the potential of a multiscalar, multiproxy site formation research to improve stratigraphic investigation of Palaeolithic sites in similar geomorphological contexts.

We study two possible cosmological consequences of a first-order phase transition in the temperature range of $1$ GeV to $10^3$ TeV: the generation of a stochastic gravitational wave background (SGWB) within the sensitivity of the Laser Interferometer Space Antenna (LISA) and, simultaneously, primordial magnetic fields that would evolve through the Universe's history and could be compatible with the lower bound from $\gamma$-ray telescopes on intergalactic magnetic fields (IGMF) at present time. We find that, if even a small fraction of the kinetic energy in sound waves is converted into MHD turbulence, a first order phase transition occurring at temperature between $1$ and $10^6$ GeV can give rise to an observable SGWB signal in LISA and, at the same time, an IGMF compatible with the lower bound from the $\gamma$-ray telescope MAGIC, for all proposed evolutionary paths of the magnetic fields throughout the radiation dominated era. For two values of the fraction of energy density converted into turbulence, $\varepsilon_{\rm turb}=0.1$ and $1$, we provide the range of first-order phase transition parameters (strength $\alpha$, duration $\beta^{-1}$, bubbles wall speed $v_w$, and temperature $T_*$), together with the corresponding range of magnetic field strength $B$ and correlation length $\lambda$, that would lead to the SGWB and IGMF observable with LISA and MAGIC. The resulting magnetic field strength at recombination can also correspond to the one that has been proposed to induce baryon clumping, previously suggested as a possible way to ease the Hubble tension. In the limiting case $\varepsilon_{\rm turb} \ll 1$, the SGWB is only sourced by sound waves, however, an IGMF is still generated. We find that values as small as $\varepsilon_{\rm turb} \sim O(10^{-13})$ (helical) and $O (10^{-9})$ (non-helical) can provide IGMF compatible with MAGIC's lower bound. International audience

Abstract A collaboration between the Linked Art II project and Participatory Knowledge Practices in Analogue and Digital Image Archives (PIA) has led to a reconfigurable Python-based workflow to transform cultural heritage data, initially photographic collections, into Linked Open Usable Data (LOUD), as a foundation for varied participatory interfaces supporting scholarship and beyond. Motivation PIA, led by the University of Basel and the Bern Academy of the Arts, aims to encourage participation from scholars and the wider public through three collections from the photographic archives of the Swiss Society for Folklore Studies (SSFS): The Atlas of Swiss Folklore, Ernst Brunner, and Kreis Family. PIA aims to create multiple interfaces reflecting diverse perspectives by deploying community-developed LOUD specifications such as the International Image Interoperability Framework (IIIF) and Linked Art, an RDF application profile (JSON-LD) based on CIDOC-CRM to describe object-based cultural heritage. In collaboration with the University of Oxford via the Linked Art II project, PIA has transformed their cultural heritage collection data into Linked Art using templates, which encapsulate the data characteristics and cataloguing practices. A Linked Art API will provide an additional entry point, as a means of conveying semantically enriched events and as a benchmark against other collections leveraging this model. Python-based Workflow To generate Linked Art files for the combined PIA collection a data transformation workflow has been created through which PIA templates were encoded in Python for a given Linked Art API entity endpoint, currently: DigitalObject HumanMadeObject and Set. The workflow, described in a use case example with a photograph from the Ernst Brunner collection, provides a three-step software process for transforming data into Linked Art. Query: The first Python script extracts collection data from the (legacy) PIA JSON API. YAML front matters are used for script variables. The filepath for the relevant .yaml file is specified as a script argument throughout the workflow. For our example, the script queried data for all images looking for DigitalObject entities (one of the variables) and the object's metadata that are stored come specifically from https://json.participatory-archives.ch/api/v1/images/12033} Map: Templates are used to map collection data to an intermediate JSON data format. The intermediate JSON data format means that the later transformation script that creates Linked Art JSON-LD does not necessarily need to be modified if a new data source is introduced. Transform: The intermediate JSON data format is transformed to Linked Art with Python functions that define ‘patterns’ (for example classified_as) for representing different aspects of photographic collection data as Linked Art. Future Work The SSFS will migrate its database into the DaSCH Service Platform (DSP) and amend their data model which will affect PIA, requiring an upgrade of its infrastructure and APIs. The workflow for creating Linked Art representations will have to be reconfigured and repurposed with different data sources. After the migration of the SSFS database, the PIA team will investigate remaining issues in the GitHub repository regarding the correct mapping of Linked Art entities and attribution of IDs. The team will work to ensure that the appropriate Linked Art modelling is achieved through the workflow. Conclusion The reconfigurable Python-based workflow is able to transform cultural heritage data, initially photographic collections, into LOUD, as a foundation for varied participatory interfaces supporting scholarship and beyond. The adaptability and extensibility of the workflow allows for potential future transformations of data from other collections to Linked Art. Acknowledgements This work has been supported by the PIA research project which is funded by the Swiss National Science Foundation (SNSF) as part of the Sinergia funding scheme (contract no. CRSII5_193788) and by the Linked Art II project at the University of Oxford (Principal Investigator: Dr. Kevin R. Page, Oxford e-Research Centre) funded by the UK Arts and Humanities Research Council (AHRC project reference AH/T013117/1). {"references": ["Newbury, David (2018): 'LOUD: Linked Open Usable Data and linked.art', in 2018 CIDOC Conference. CIDOC Annual Conference, Heraklion, Greece: International Council of Museums, pp. 1\u201311. [01.04.2023].", "Page, Kevin R. / Delmas-Glass, Emmanuelle / Beaudet, David / Norling, Samantha / Rother, Lynn / H\u00e4nsli, Thomas (2020): 'Linked Art: Networking Digital Collections and Scholarship', in DH2020 Book of Abstracts. Digital Humanities 2020, Online, pp. 504\u2013509. [01.04.2023].", "Raemy, Julien Antoine / Demleitner, Adrian (2023): 'Implementation of the IIIF Presentation API 3.0 based on Software Support: Use Case of an Incremental IIIF Deployment within a Citizen Science Project', in Digital Heritage. Progress in Cultural Heritage: Documentation, Preservation, and Protection. Cham: Springer International Publishing (Lecture Notes in Computer Science). Unpublished paper presented at the International Conference on Digital Heritage, Limassol, Cyprus, November 7-11, 2022.", "Sanderson, Robert (2018): 'Shout it Out: LOUD'. EuropeanaTech Conference 2018, Rotterdam, the Netherlands, 15 May. [01.04.2023].", "Sanderson, Robert (2019): 'Keynote: Standards and Communities: Connected People, Consistent Data, Usable Applications', in 2019 ACM/IEEE Joint Conference on Digital Libraries (JCDL). 2019 ACM/IEEE Joint Conference on Digital Libraries (JCDL), Urbana-Champaign, Illinois, USA: IEEE, p. 28. DOI: 10.1109/JCDL.2019.00009."]} This poster is related to the extended abstract that is published as part of the DH2023 Book of Abstracts. All versions are identical, only the malfunctioning QR code was replaced.