Central European transformation is illustrated here through everyday scenes of border traffic in the Mukkerman chapter of Ádám Bodor’s novel Schutzgebiet Sinistra (1992). The narrated transit space serves as a symbolic representation of a potentially imminent societal transformation process. Various motifs cast both sides – or the actors of transformation (and those resisting it) – and their narratives in an ironic light. The analysis pays particular attention to the linguistic perforation and subversion of sovereign practices of the border regime within a fictional zone in the Romanian-Hungarian-Ukrainian borderlands during the 1980s.
Lietuvių spontaninės kalbos automatinis atpažinimas išlieka sudėtinga užduotis dėl ribotų išteklių, tarimo variacijų ir nuostolingo garso įrašų poveikio transkribavimo kokybei. Šio straipsnio tikslas – ištirti Whisper large‑v3 modelio pritaikymą lietuvių spontaninės kalbos atpažinimui taikant Low‑Rank Adaptation (LoRA) strategiją, palyginti jį su baziniu modeliu ir išanalizuoti, kaip papildomas apmokymas veikia transkribavimo kokybę. Gauti rezultatai parodė, kad LoRA strategija reikšmingai pagerino modelio veikimą ir sumažino tiek tekstinį, tiek fonetinį nutolimą nuo tikrųjų transkripcijų.
Modern software systems increasingly rely on APIs, making efficient testing essential for ensuring reliability and correctness. However, creating system tests from natural language requirements is typically a manual and time-consuming process. This research proposes an approach for automatically generating executable REST API tests from natural language requirements using large language models (LLMs) and OpenAPI specifications. A prototype system was developed and evaluated on two real-world APIs. The results show that the approach can generate executable pytest tests, achieve full endpoint reachability coverage, and provide stable automated test execution.
After 1991 the vulnerability of the working class staring at the loss of their livelihood and increasing precarity has been addressed less and less by the left parties and unions in Ukraine. Consequently, working-class discontent with neoliberal capitalism increasingly became channelled into the support of far-right political forces. The left parties established in Ukraine in the early 1990s failed to speak on behalf of the wage workers. In the 2000s some of them disappeared from politics (SPU and SDPU), turned into minor allies of mainstream neoliberal parties (CPU), or were marginalized and instrumentalized by Russian intelligence service (PSPU). An absence of the left alternative in the post-socialist state captured by the oligarchs and corrupted officials led to the growing resentment in the industrial regions accompanied by the fears of further deindustrialization. Russia masterly used these fears and resentment to fuel the secessionist insurgency in Donbas in spring 2014. Collapse of the ruling Party of the Regions in spring 2014 created political vacuum in Donbas which was filled by Russian nationalists, the Don Cossacks and their local supporters – not numerous but quite aggressive to fuel the insurgency in Donbas. Local members of the quasi-left CPU and PSPU – particularly in Luhansk region – actively supported the right-wing insurgents in spring-summer 2014. Since then, Donbas came through accelerated decline of its industries and final deindustrialization due to the Russian invasion of 2022.
This paper proposes a graphRAG-driven method for validating the semantic, content, and data compliance of software requirements against organisational and technical reference documentation, using a graph-based retrieval-augmented generation system paired with a large language model. The experimental evaluation demonstrated that the proposed method achieved high semantic compliance accuracy in its baseline graphRAG configuration, while adjusting graph database parameters and prompting techniques yielded improvements in requirements compliance accuracy.
A certain paradox has pervaded German-Baltic relations for decades. Tightly linked through culture, art, education, and, above all, a genuine societal affinity for each other, in the realm of military power, the two sides could not be further apart. Where, over the years, German leadership has emphasized the notion of perpetual peace in Europe and positive transformation through trade, the Baltic states, on the contrary, stuck to the worldview of military power playing a central role in regional affairs. Put differently, despite geographic proximity between Germany and the Baltic states, these relations time and again have fractured along the line of how to best engage Russia. Existing scholarship has already extensively covered Germany’s security ties with other major powers. This essay, in turn, seeks to depart from this conventional practice and instead assess Germany as a security actor through the eyes of small states in the Baltic Sea region, namely Estonia, Latvia, and Lithuania. The following traces and evaluates key inflection points and controversies nestled within the German-Baltic security relationship.
This paper compares Random Forest classification and regression approaches for equity trading strategy development using Zigzag-based price labeling. Both models are trained to identify buy and sell signals derived from Zigzag pivot points and are evaluated against a buy-and-hold benchmark using three exchange-traded funds: S&P 500, Hang Seng, and MSCI UK. A walk-forward framework is implemented using five-year training windows and two-year out-of-sample test windows. The results indicate that the Random Forest classification strategy delivers the strongest performance, particularly in the Hong Kong market, and is found to statistically significantly outperform the regression strategy. Further analysis of pivot timing shows that both models exhibit higher accuracy in predicting buy signals than sell signals, suggesting an asymmetry in model effectiveness across market turning points. As Zigzag labels are retrospective by construction, thus results should be interpreted as an upper bound on achievable performance rather than a direct estimate of live trading returns.
Elektroninės komercijos svetainėse vis dažniau pasitaiko vartotojo sąsajos sprendimų, kurie gali pakreipti vartotojus paslaugos tiekėjams palankia linkme ir taip daryti įtaką jų pasirinkimams. Tokie sprendimai, vadinami apgaulingais šablonais, vis dažniau yra laikomi nelegalia praktika, todėl svarbu gebėti atpažinti šiuos šablonus ir įvertinti jų poveikį vartotojams. Straipsnyje siekiama parengti vertinimo euristikas ir metodiką priemonei, kuri padėtų atpažinti apgaulingus sąsajos elementus. Tyrimo metu buvo išskirti etiško projektavimo principų pažeidimų požymiai, o pagal apgaulingų šablonų ir etiško projektavimo principų pažeidimų sąsajas buvo nustatytos vertinimo euristikos. Sukurta metodika skirta padėti ne tik atpažinti apgaulingus šablonus, bet ir nustatyti svetainės etiško projektavimo principų pažeidimų profilį, kuris nurodo galimą neigiamą poveikį naudotojui, taip prisidedant prie etiškesnių skaitmeninių sistemų projektavimo skatinimo.
This study evaluates the ability of lightweight open-source large language models (LLMs) to detect bias in text. Eleven models of six popular LLM families were tested in a zero-shot setting on a unified dataset of 8,745 sentences derived from three selected sources, covering gender, race, religion, and appearance bias. Results showed that none of the models exceeded 70% accuracy, which highlighted limitations of lightweight LLMs and existing challenges related to current bias detection datasets.
Šiame darbe nagrinėjama, ar akcijų kainos kyla prieš įmonių įtraukimą į S&P500 indeksą (angl. Standard and Poor’s 500 index) ir ar tai galima panaudoti investavimo strategijoje. Analizuojami 2021–2026 m. duomenys, vertinant kainų pokyčius 30, 60 ir 90 dienų iki įtraukimo. Rezultatai rodo, kad akcijų kainos dažnai pradeda kilti dar prieš oficialų paskelbimą, o ilgesnis investavimo laikotarpis siejamas su didesne grąža. Teoriniame scenarijuje galima pasiekti apie 59 % vidinę metinę grąžos normą (angl. Extended Internal Rate of Return, XIRR) investuojant prieš 30 dienų, apie 84 % prieš 60 dienų ir apie 90 % prieš 90 dienų. Vis dėlto praktikoje tokia strategija yra ribota dėl prognozavimo netikslumų, tačiau rezultatai rodo, kad ši kryptis turi potencialo tolesniems tyrimams.
The introduction outlines the theme of the conference and the conditions under which the volume emerged against the backdrop of the geopolitical changes of recent years. The key concepts of “transformation” and “narrative” are examined in their interdisciplinary dimensions. Finally, the five sections and the individual contributions are briefly introduced.
In Central and Eastern Europe, the National Socialist and Communist dictatorships have often been equated as ‘totalitarian’ regimes memory politics from 1989-91 onwards. While the cooperation with the Nazi occupiers in the Second World War have been was largely (though by no means universally) ignored or even glorified, many post-Soviet governments have endorsed and disseminated nationalist memory narratives. By contrast, Vladimir Putin has justified the Molotov-Ribbentrop Pact of 23 August 1989 and the ensuing occupation of East-Central European states as defensive security measures. As opposed to the post-Soviet and nationalist narratives, memory politics in Western and Central Europe (as well as in the United States) has highlighted the mass murder of the Jews since the 1990s. The article outlines these tensions, which prevent a common European culture of remembrance from emerging in the foreseeable future.
Didėjant e. komercijos sistemų apkrovoms, duomenų bazių našumas ir mastelio plėtimas tampa kritiniu veiksniu užtikrinant sklandžią naudotojų patirtį. Šiame darbe sprendžiama problema – kuri duomenų bazių valdymo sistema yra efektyvesnė aukšto srauto scenarijuose: rakto–reikšmės tipo Redis Cluster ar atmintyje veikianti reliacinė PostgreSQL su Citus plėtiniu. Tyrimo rezultatai rodo, kad Redis Cluster pasiekia žymiai didesnį pralaidumą net ir tada, kai PostgreSQL veikia vienodomis atminties sąlygomis, tačiau PostgreSQL pasižymi stabilesne delsa esant mažesnėms apkrovoms.
This study investigates the performance of the VITS model for Lithuanian speech synthesis under different training configurations. Experiments were conducted using datasets with phoneme-based and grapheme-based text representations, accented text, and both single-speaker and multi-speaker setups. The goal was to evaluate how linguistic pre-processing and speaker diversity influence synthesis quality. Model outputs were compared using objective measures. The results provide insights into the impact of phoneme representation and accent information on the quality of Lithuanian neural TTS systems.
Vartotojų praradimo identifikavimas leidžia programėles valdančioms įmonėms įvertinti įvairų rodiklių įtaką klientų išlaikymui ir planuoti artėjančią tinklų apkrovą. Šiame tyrime analizuojami vartotojų aktyvumo bei programėlės techninės kokybės duomenys ir taikomi statistiniai bei mašininio mokymosi modeliai siekiant įvertinti, kurie vartotojai atsisakys paslaugos pasibaigus abonementui. Rezultatai rodo, kad tinkamiausias modelis nagrinėjamo duomenų rinkinio klasifikavimui – atsitiktiniai miškai.
Šiame tyrime nagrinėjamas valandinis saulės fotovoltinių elektrinių energijos prognozavimas Lietuvoje, lyginant tiesioginio ir dviejų pakopų prognozavimo strategijas. Analizei taikyti kelių grupių modeliai: klasikiniai mašininio mokymosi modeliai – atraminių vektorių modelis (SVM), atsitiktinių miškų modelis (RF) ir ekstremalaus gradientinio stiprinimo modelis (XGBoost), bei gilieji neuroniniai tinklai – ilgos trumpalaikės atminties (LSTM) ir laikinis suliejimo transformeris (TFT). Tiesioginio prognozavimo strategijoje (kai modelis tiesiogiai prognozuoja energijos gamybą, naudodamas meteorologinius duomenis) geriausi dieninės paklaidos rezultatai gauti naudojant XGBoost modelį šiaurės ir centrinėje elektrinėse, kur dienos svorinė vidutinė absoliuti procentinė paklaida (DWMAPE) siekė atitinkamai 13,24 % ir 16,25 %, o pietinėje elektrinėje geriausius rezultatus pasiekė TFT modelis, pasiekęs 12,68 % DWMAPE. Dviejų pakopų strategijos pirmoje pakopoje bendros dienos energijos prognozėms tiksliausi modeliai skyrėsi tarp elektrinių: šiaurės elektrinėje geriausius rezultatus pasiekė RF modelis, centrinėje – SVM, o pietinėje – TFT. Antroje pakopoje valandinių proporcijų prognozėse visose elektrinėse geriausi rezultatai gauti naudojant RF modelį. Galutiniai rezultatai parodė, kad tiesioginio prognozavimo strategija visose trijose elektrinėse buvo tikslesnė už dviejų pakopų strategiją. Tiesioginio prognozavimo strategijoje XGBoost pasižymėjo mažiausia rezultatų sklaida (σ = 1,56) pagal DWMAPE tarp visų elektrinių. Dviejų pakopų prognozavimo strategijos pirmoje pakopoje mažiausia rezultatų sklaida pasižymėjo SVM modelis (σ = 1,66), o antroje pakopoje – XGBoost modelis (σ = 1,12).
Šiame straipsnyje tiriamas požymių atrankos ir eliminavimo poveikis studentų studijų nutraukimo prognozavimo kokybei. Duomenų rinkinį sudaro skirtingų grupių požymiai: akademiniai, demografiniai, instituciniai, socioekonominiai, elgsenos ir išvestiniai elgsenos. Atliekama požymių svarbos analizė ir nuoseklus jų mažinimas etapais, siekiant nustatyti optimalų požymių rinkinį. Rezultatai rodo, kad didžiausią įtaką turi elgsenos dinamika ir jos išvestiniai rodikliai, o perteklinių požymių šalinimas supaprastina modelį, tačiau kartu mažina jo gebėjimą tiksliai identifikuoti studijų nutraukimo atvejus.
This article is dedicated to analyzing public criminological discourse in Stalinist, post-Stalinist and post-Soviet Lithuania. Its chronological boundaries are between 1940 (when Soviet first-time occupied country) and early 2000s (when Lithuania released new, not Soviet-based criminal code, finalized post-Soviet criminal prosecution reform and joined European Union and NATO). The article does not include the period of Nazi Germany occupation of Lithuania (1941–1944), since the conditions of that period are very different and requires additional research. Early 2000s have also marked the end of mass crime and gang and mafia violence. Since then, crime numbers considerably decreased and today are relatively low. The public discourse, as the article reveals, does not necessarily match the crime reality. For instance, in Stalinist period, the politically constructed enemy was the dominating definition of the criminal.
Autism spectrum disorder affects communication abilities, making augmentative and alternative communication systems essential for nonverbal individuals. The Picture Exchange Communication System is a structured, symbol-based method that enables nonverbal individuals to initiate communication by exchanging picture cards representing desired objects, actions, or concepts. It is widely used but requires efficient card selection strategies to maximize successful exchanges. This research evaluates the effectiveness of reinforcement learning algorithms combined with different card selection strategies. The experiments revealed that introducing opposite card selection after rejection significantly improved success rate compared to similarity-only approaches. The findings demonstrate that strategy design has greater impact on system performance than the choice of learning algorithm.
Until the early 1990s, in reunified Germany, the prevailing view was that deploying the Bundeswehr abroad could not be justified for political and, above all, historical reasons. It was only the ruling of the Federal Constitutional Court on July 12, 1994, that made it possible for German armed forces to be deployed abroad under a NATO or UN mandate, but only with the approval of the German Bundestag. Since then, the Bundeswehr has been deployed abroad on several occasions, including in Afghanistan from 2002 to 2021, in Iraq from 2014 to 2017, and currently in Mali. However, its operational deployment has been and continues to be characterized primarily by military restraint and a focus on maintaining order and providing civilian reconstruction assistance. Compared to its military deployment, reunified Germany has shown much greater intensity in the field of European policy, particularly with the aim of furthering the integration of the EC/ EU member states. This policy has been limited by four major international crises (the global financial and economic crisis of 2008–2011; the euro crisis from 2012–2015; the refugee crisis since 2015/16; and the COVID-19 pandemic since February 2020), slowing down the further integration of EU states. During these years, Germany has only partially succeeded in transferring its leadership role in the EU into greater capacity for action, as the Federal Republic’s multilateral approach is not always supported by its European partners.
