Our Work

This Introduction explores the resources of a modern intellectual, the writer Richard Sakwa for an analysis of the current role and significance of an author as political figure. Suggesting that the t…

This Introduction explores the resources of a modern intellectual, the writer Richard Sakwa for an analysis of the current role and significance of an author as political figure. Suggesting that the tradition of political science and political sociology needs new approaches and inspirations from other disciplines that take their horizon broader. This helps to explain Sakwa's inspiration from political anthropology, from Girard's mimetic theory, from political philosophy of Deleuze and Guattari, when he creatively re-developed them into his life work. In honour of Richard Sakwa we regard the complexity of a lifework that is exercised in a more authentic way.

This record contains the code accompanying the article "Skillful Data-Driven Subseasonal Soil Moisture Forecasting: Prospects and Limits for Flash Drought Prediction" (Otero, Özer, Fernánd…

This record contains the code accompanying the article "Skillful Data-Driven Subseasonal Soil Moisture Forecasting: Prospects and Limits for Flash Drought Prediction" (Otero, Özer, Fernández-Torres, Ma). The repository includes:- the SMCast model and the deep-learning baselines used in the paper (UNet, E3DLSTM, DroughtFormer), all at a similar parameter count;- training scripts (deterministic training and probabilistic quantile-head fine-tuning);- evaluation scripts that generate predictions and metrics, including persistence and climatology reference forecasts, plus aggregation of metrics across training seeds;- preprocessing scripts that turn raw ERA5 into model inputs, soil-moisture anomalies, flash-drought indices, percentile thresholds and train/validation/test splits. Dependencies are managed with uv (Python 3.10; the lockfile pins CUDA 12.8 PyTorch wheels). ERA5 data are not included and are available from the Copernicus Climate Data Store. Data paths are set in the YAML files under config/. See README.md for installation, data setup, training and evaluation commands. Released under the MIT License.

Slides of an online presentation regarding Wikidata organized by Babeș-Bolyai University (17 November 2025). The recording is available on Youtube: https://www.youtube.com/watch?v=4iiL9WFx2JY Slides w…

Slides of an online presentation regarding Wikidata organized by Babeș-Bolyai University (17 November 2025). The recording is available on Youtube: https://www.youtube.com/watch?v=4iiL9WFx2JY Slides were originally uploaded at: https://commons.wikimedia.org/wiki/File:An_introduction_to_Wikidata_and_its_use_for_librarians_and_researchers.pdf

Diffusion language models (DLMs) generate multiple tokens with a single forward pass per decoding round, achieving greater parallelism than autoregressive models (ARMs). Revision promises further redu…

Diffusion language models (DLMs) generate multiple tokens with a single forward pass per decoding round, achieving greater parallelism than autoregressive models (ARMs). Revision promises further reductions in sequential computation by allowing a DLM to overwrite previously generated tokens and thereby reducing the number of decoding rounds. Yet it remains unclear whether this advantage persists as DLMs evolve. To assess the expressive power of DLMs with and without revision, we work in a model where each Transformer's decoding round uses $\mathrm{AC}^0$ computation. We compare the minimum number of rounds each class needs to sample $(X,f_L(X))$, where the word $X$ is uniform on $\{0,1\}^n$ and $f_L$ is the membership function of a binary regular language $L$. We show that revision yields an asymptotic advantage. A DLM with revision exactly samples this distribution in at most two rounds without chain-of-thought (CoT) positions. However, a DLM without revision has tight round complexity $\Theta(\log n/\log\log n)$ even with a constant number of CoT positions for every $L\notin\mathrm{AC}^0$, i.e., every regular language whose membership cannot be computed by such circuits. Moreover, for every such $L$ and every constant $\theta>0$, $O(n^{1-\theta})$ CoT positions do not suffice to sample this distribution in constant rounds without revision. We further show that there is a separation when the goal is to approximately sample a length-$n$ word uniformly from some regular language $L$. These results provide a theoretical framework for understanding how revision reduces sequential computation and expands the scope of parallel speedups achievable by DLMs.

Col-detection and analysis code reproducing the figures and reported quantities of Paul (2026, Weather and Climate Dynamics). Includes a manifest linking each figure and result to its producing script…

Col-detection and analysis code reproducing the figures and reported quantities of Paul (2026, Weather and Climate Dynamics). Includes a manifest linking each figure and result to its producing script.

A educação inclusiva constitui-se como uma diretriz fundamental das políticas públicas educacionais brasileiras, estando diretamente relacionada à garantia do direit…

A educação inclusiva constitui-se como uma diretriz fundamental das políticas públicas educacionais brasileiras, estando diretamente relacionada à garantia do direito à educação, à equidade e à justiça social. O estudo tem como objetivo analisar os desafios e as possibilidades da gestão escolar na implementação da educação inclusiva na rede pública de ensino, destacando o papel do gestor escolar como agente estratégico da gestão pública. A pesquisa caracteriza-se como bibliográfica, de abordagem qualitativa, fundamentada em livros, artigos científicos, legislações e documentos oficiais que abordam a educação inclusiva, a gestão escolar e as políticas públicas educacionais. Os resultados evidenciam que a efetivação da inclusão escolar depende de fatores como formação continuada dos profissionais, planejamento pedagógico inclusivo, gestão democrática e articulação intersetorial. Conclui-se que, apesar das dificuldades estruturais e institucionais, a gestão escolar possui possibilidades concretas de atuação para promover uma educação inclusiva e de qualidade na rede pública.

Batch upload of files from directory 'output'.

Replication package for the paper 'The Price of Asking: preference elicitation under time pressure for dynamic flexible job-shop rescheduling'. It contains the complete source code, the benchmark inst…

Replication package for the paper 'The Price of Asking: preference elicitation under time pressure for dynamic flexible job-shop rescheduling'. It contains the complete source code, the benchmark instances, the pre-generated candidate pools, the simulation and analysis scripts, the code and stored outputs of every revision campaign, and the anonymized data and applications of the two behavioral studies (pairwise pilot and industrial questionnaire).

Periodic P–T–Kzz synchronization was added to improve integration stability after we found cases where continuous synchronization caused oscillations and prevented convergence. Periodic mode holds tem…

Periodic P–T–Kzz synchronization was added to improve integration stability after we found cases where continuous synchronization caused oscillations and prevented convergence. Periodic mode holds temperature and eddy diffusion fixed between resynchronizations, then requires chemistry to reconverge after updates. Highlights Gas giant initialization defaults to periodic synchronization (mode=0). Continuous synchronization (mode=1) remains available and is still the base EvoAtmosphere default. Periodic profiles and TOA maintenance require the robust stepper. Robust integration coordinates TOA and profile maintenance, handles extreme mismatches immediately, and permits recoverable measurement/resync failures through a shared retry limit. Gas giant save/restore preserves the exact installed profile and its settings, validates inputs, and copies saved arrays independently. It does not restore solver history or integration time. API compatibility Pressure-profile initialization accepts integer mode, followed by hydro_pressure. Prefer keyword arguments for optional settings; scalar defaults generally replace None. TOA-specific failure and update-interval controls are replaced by var.max_resync_failures and var.nsteps_before_reinit. Shared diagnostics include wrk.n_resync_failures and wrk.nconverged_but_restarted. Older gas giant dictionaries missing required state fields are rejected.

MeasureIA is a tool that can be used to easily measure intrinsic alignment correlation functions and clustering in simulation boxes and lightcone data.