Back to Search
Start Over
Collectible Card Games cards generation with neural network
- Publication Year :
- 2022
- Publisher :
- СанкÑ-ÐеÑеÑбÑÑгÑкий полиÑÐµÑ Ð½Ð¸ÑеÑкий ÑнивеÑÑиÑÐµÑ ÐеÑÑа Ðеликого, 2022.
-
Abstract
- ÐейÑоннÑе ÑеÑи и ÑеÑи Ð´Ð»Ñ Ð³ÐµÐ½ÐµÑаÑии ÑекÑÑа, Ñакие как GPT-2, заÑабоÑали болÑÑÑÑ Ð¿Ð¾Ð¿ÑлÑÑноÑÑÑ Ð² ÑазнÑÑ Ð¾Ð±Ð»Ð°ÑÑÑÑ Ð¿ÑименениÑ. ÐейÑоннÑе ÑеÑи Ñакого Ñипа ÑаÑÑо иÑполÑзÑÑÑÑÑ Ð¸ÑÑледоваÑелÑми и ÑнÑÑзиаÑÑами Ð´Ð»Ñ Ð¿Ð¾Ð»ÑÑÐµÐ½Ð¸Ñ Ð¾ÑвеÑов на заданнÑе вопÑоÑÑ Ð¸Ð»Ð¸ Ð´Ð»Ñ ÑÐ¾Ð·Ð´Ð°Ð½Ð¸Ñ ÑекÑÑов по опÑеделÑÐ½Ð½Ð¾Ð¼Ñ ÑоÑмаÑÑ. РпÑимеÑÑ, Ñакие модели могÑÑ Ð¸ÑполÑзоваÑÑÑÑ Ð¿Ñи Ñоздании игÑ. Ð ÑÐ°Ð¼ÐºÐ°Ñ ÐºÐ²Ð°Ð»Ð¸ÑикаÑионной ÑабоÑÑ Ñеализован алгоÑиÑм дополниÑелÑного обÑÑÐµÐ½Ð¸Ñ Ð¼Ð¾Ð´ÐµÐ»ÐµÐ¹ генеÑаÑии ÑекÑÑа пÑи помоÑи генеÑаÑивно-ÑоÑÑÑзаÑелÑного меÑода. Ðдна из Ñелей ÑабоÑÑ â опÑеделиÑÑ, ÑпоÑÐ¾Ð±Ð½Ñ Ð»Ð¸ модели дополниÑелÑно обÑÑиÑÑÑÑ Ð¾ÑобенноÑÑÑм даннÑÑ Ð¿Ñи иÑполÑзовании данного меÑода. РбÑдÑÑем меÑод Ð¼Ð¾Ð¶ÐµÑ Ð±ÑÑÑ ÑлÑÑÑен и модиÑиÑиÑован. Ðнализ ÑÑÑекÑивноÑÑи моделей бÑÐ´ÐµÑ Ð¿ÑоводиÑÑÑÑ Ð½Ð° оÑновании иÑÑÐ»ÐµÐ´Ð¾Ð²Ð°Ð½Ð¸Ñ Ð¾Ð¿ÑоÑа гÑÑÐ¿Ð¿Ñ Ð¿Ð¾Ð»ÑзоваÑелей, коÑоÑÑм бÑла поÑÑавлена задаÑа на вÑделение лÑÑÑего вÑвода ÑеÑи. РкаÑеÑÑве оÑновной модели, в ÑабоÑе иÑполÑзÑеÑÑÑ GPT-2. ÐÐ°Ð½Ð½Ð°Ñ Ð¼Ð¾Ð´ÐµÐ»Ñ Ð¸ÑполÑзÑеÑÑÑ ÐºÐ°Ðº в каÑеÑÑве генеÑаÑоÑа, Ñак и в виде клаÑÑиÑикаÑоÑа в Ñоли диÑкÑиминаÑоÑа в генеÑаÑивно-ÑоÑÑÑзаÑелÑной модели. ÐÑе модели бÑли поÑÑÑÐ¾ÐµÐ½Ñ Ð¸ обÑÑÐµÐ½Ñ Ð¿Ñи помоÑи pyTorch. ТеÑÑÐ¾Ð²Ð°Ñ ÑÑÑаниÑа в ÑеÑи ÐнÑеÑÐ½ÐµÑ Ð±Ñла Ñоздана пÑи помоÑи Flask.<br />Neural networks, and text generation models like GPT-2, gained popularity in various applications. Text generation networks usually help researchers and enthusiasts to answer on given questions or to generate text like some style or format. For example, they can be used to generate ideas for games. In this paper will be covered complex algorithm of model training using standard fine-tuning and Generative Adversarial method. One objective is to evaluate, is this method can be used for text generation tuning and identify potential problems. In the future this algorithm can be modified further, to fit problems better. An analysis of algorithm effectivity will be evaluated with group of users that will be tasked to select most fitting text for given problem. In this paper GPT-2 was used as main model. GPT-2 language model was used for text generation and GPT-2 for sequence classification was used as discriminator in GAN method. All models were trained with pyTorch. Testing website was made with Flask.
Details
- Language :
- Russian
- Database :
- OpenAIRE
- Accession number :
- edsair.doi...........f928c297eb4e4aeb75574cf347d10799
- Full Text :
- https://doi.org/10.18720/spbpu/3/2022/vr/vr22-3178