[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"detail-sidebar-cat-0-en-105":3,"doc-seo-502508-105":59,"doc-detail-502508-en":130},{"code":4,"msg":5,"data":6},0,"success",[7,13,18,23,28,33,38,43,48,51,55],{"id":8,"doc_module":4,"doc_module_name":9,"category_name":10,"show_sort_weight":11,"slug":12},1,"Document","Story & Novel",90,"story-novel",{"id":14,"doc_module":4,"doc_module_name":9,"category_name":15,"show_sort_weight":16,"slug":17},2,"Literature",80,"literature",{"id":19,"doc_module":4,"doc_module_name":9,"category_name":20,"show_sort_weight":21,"slug":22},4,"Exam",70,"exam",{"id":24,"doc_module":4,"doc_module_name":9,"category_name":25,"show_sort_weight":26,"slug":27},5,"Comic",60,"comic",{"id":29,"doc_module":4,"doc_module_name":9,"category_name":30,"show_sort_weight":31,"slug":32},6,"Technology",50,"technology",{"id":34,"doc_module":4,"doc_module_name":9,"category_name":35,"show_sort_weight":36,"slug":37},7,"Healthcare",40,"healthcare",{"id":39,"doc_module":4,"doc_module_name":9,"category_name":40,"show_sort_weight":41,"slug":42},8,"Research & Report",30,"research-report",{"id":44,"doc_module":4,"doc_module_name":9,"category_name":45,"show_sort_weight":46,"slug":47},9,"Religion & Spirituality",20,"religion-spirituality",{"id":46,"doc_module":4,"doc_module_name":9,"category_name":49,"show_sort_weight":46,"slug":50},"World Cup","world-cup",{"id":52,"doc_module":4,"doc_module_name":9,"category_name":53,"show_sort_weight":52,"slug":54},10,"Lifestyle","lifestyle",{"id":56,"doc_module":4,"doc_module_name":9,"category_name":57,"show_sort_weight":24,"slug":58},19,"General","general",{"code":4,"msg":60,"data":61},"ok",{"site_id":62,"language":63,"slug":64,"title":65,"keywords":66,"description":67,"schema_data":68,"social_meta":123,"head_meta":125,"extra_data":127,"updated_unix":129},105,"en","dynamic-jointly-batch-selection-for-data-efficient-machine-translation-fine-tuning","Dynamic Jointly Batch Selection for Data Efficient Machine Translation Fine-Tuning","","Data quality and effective selection underpin strong machine translation performance and robust, reliable translation systems. This paper introduces a fine-tuning data selection methodology that combines a learner model with a pre-trained reference model to improve training effectiveness. It defines a learnability score to score each data point’s utility and uses a batch selection strategy that accounts for interdependencies among points. Experiments on English↔Persian and multiple other pairs with mBART on CCMatrix show up to fivefold data-efficiency gains over an iid baseline and 24% computational-efficiency improvement using cached embeddings. It also improves generalization beyond random selection.",{"@graph":69,"@context":122},[70,84,105],{"@type":71,"itemListElement":72},"BreadcrumbList",[73,77,79,82],{"item":74,"name":75,"@type":76,"position":8},"https://docshare.wps.com","Home","ListItem",{"item":78,"name":9,"@type":76,"position":14},"https://docshare.wps.com/document/",{"item":80,"name":40,"@type":76,"position":81},"https://docshare.wps.com/document/research-report/",3,{"item":83,"name":65,"@type":76,"position":19},"https://docshare.wps.com/document/dynamic-jointly-batch-selection-for-data-efficient-machine-translation-fine-tuning/502508/",{"url":83,"name":65,"@type":85,"image":86,"author":91,"headline":65,"publisher":94,"fileFormat":97,"inLanguage":63,"description":67,"dateModified":98,"datePublished":99,"encodingFormat":97,"isAccessibleForFree":100,"interactionStatistic":101},"DigitalDocument",{"url":87,"@type":88,"width":89,"height":90},"https://docshare.wps.com/thumbnails/dynamic-jointly-batch-selection-for-data-efficient-machine-translation-fine-tuning/502508.png","ImageObject",300,407,{"name":92,"@type":93},"Putri","Person",{"url":74,"name":95,"@type":96},"DocShare","Organization","application/pdf","2026-10-16","2026-10-03",true,{"@type":102,"interactionType":103,"userInteractionCount":24},"InteractionCounter",{"@type":104},"ViewAction",{"@type":106,"mainEntity":107},"FAQPage",[108,114,118],{"name":109,"@type":110,"acceptedAnswer":111},"What problem does the paper address in machine translation fine-tuning?","Question",{"text":112,"@type":113},"It targets improving machine translation by selecting more useful training data, since irrelevant or poorly translated pairs can degrade model performance. The paper also emphasizes making training more data-efficient.","Answer",{"name":115,"@type":110,"acceptedAnswer":116},"How does the proposed method evaluate the usefulness of training data?",{"text":117,"@type":113},"It uses a learnability score derived from the learner model and leverages a pre-trained reference model. The approach scores each data point’s utility for fine-tuning.",{"name":119,"@type":110,"acceptedAnswer":120},"What is the benefit of batch selection in the method?",{"text":121,"@type":113},"Batch selection explicitly considers interdependencies among data points, making it more effective than independently selecting top-scoring samples. The experiments show improved data efficiency and translation performance versus iid and random baselines.","https://schema.org",{"og:url":83,"og:type":124,"og:title":65,"og:site_name":95,"og:description":67},"article",{"robots":126,"canonical":83},"index,follow",{"doc_id":128,"site_id":62},502508,1791775879,{"code":4,"msg":5,"data":131},{"doc_id":128,"user_id":132,"nickname":92,"user_avatar":133,"doc_module":4,"category_id":39,"category_name":40,"doc_title":65,"doc_description":67,"doc_content":134,"file_id":135,"file_url":136,"file_type":137,"file_size":138,"view_count":24,"is_deleted":4,"is_public":8,"is_downloadable":8,"audit_status":8,"page_count":44,"language":139,"language_code":63,"site_id":62,"html_lang":63,"table_of_contents":140,"faqs":141,"seo_title":142,"seo_description":67,"update_tm":143,"read_time":144},962085571259,"https://ap-avatar.wpscdn.com/davatar_29158cc5080c5b710cf443261637dec0","Dynamic Jointly Batch Selection for Data Efficient Machine Translation Fine-Tuning  \nMohammad Amin Ghanizadeh and Mohammad Javad Dousti  \nDepartment of Electrical and Computer Engineering, College of Engineering, University of Tehran, Tehran, Iran {ghanizadeh.amin,[mjdousti}@ut.ac.ir](mjdousti}@ut.ac.ir)  \narXiv :2511 .04406v 1 [ cs .CL] 6 Nov 2025  \nAbstract  \nData quality and its effective selection are fundamental to improving the performance of machine translation models, serving as cornerstones for achieving robust and reliable translation systems. This paper presents a data selection methodology specifically designed for fine-tuning machine translation systems, which leverages the synergy between a learner model and a pre-trained reference model to enhance overall training effectiveness. By defining alearnability score, our approach systematically evaluates the utility of data points for training, ensuring that only the most relevant and impactful examples contribute to the fine-tuning process. Furthermore, our method employs a batch selection strategy which considers interdependencies among data points, optimizing the efficiency of the training process while maintaining a focus on data relevance. Experiments on English ↔ Persian and several other language pairs using an mBART model finetuned on the CCMatrix dataset demonstrate that our method can achieve up to a fivefold improvement in data efficiency compared to an iid baseline. Experimental results indicate that our approach improves computational efficiency by 24% when utilizing cached embeddings, as it requires fewer training data points. Additionally, it enhances generalization, resulting in superior translation performance compared to random selection method.  \n1 Introduction  \nMachine translation is a fundamental task in natural language processing. As with any data-driven learning task, the effectiveness of training heavily depends on the quality of the data. (Fenza et al., 2021 ; Gupta et al., 2021 ; Chen et al., 2021) In particular, parallel datasets may contain irrelevant sentence pairs or poorly translated documents, which negatively impact the performance of the final model.  \nBeyond the quality of data, the state of the learner model itself plays a crucial role in select-  \ning beneficial training data. For instance, studies have shown that data points associated with high loss on the learner model are typically those the model struggles to learn. (Bucher et al., 2016 ; Harwood et al., 2017) Allocating more computational resources to such data points, rather than to those the model has already mastered, can lead to more effective training.  \nTraining can be made more data-efficient by employing selection methods during the training process, such as those based on the loss of data points on the learner model, a pre-trained model, or a combination of both.  \nWe demonstrate that the batch-selection method is more effective than both the individual sampleselection and random selection method. More specifically, selecting data points within a batch, where the points are interdependent, is more effective than independently selecting high-scoring data points. Similar findings have also been reported in previous studies for multimodal learning (Evans et al., 2024) . Our experiments focus on 12 different directions, namely, Persian ↔ English, German ↔ English, French ↔ English, Finnish ↔ English, Arabic ↔ English and Hindi ↔ English.  \nAn mBART model (Liu, 2020) is used as the learner and a pre-trained LaBSE model (Feng et al., 2022) as the reference model. The pre-trained model is called reference model, while the model undergoing fine-tuning is called the learner model.  \nWe use features extracted from both the learner model and the pre-trained model for selecting the data during the training. We employ the learnability score (Mindermann et al., 2022) to select data points for fine-tuning.  \nAs demonstrated in our experiments, the use of the learnability score as","cbCaiblU8xs5pSHz","https://ap.wps.com/l/cbCaiblU8xs5pSHz","pdf",792995,"English","# Introduction\n## Learner model and learnability\n# Related Work\n## Offline data selection\n## Online data selection\n## Hard negative mining\n## Batch selection\n# Methodology\n## Selection criteria","[{\"question\":\"What problem does the paper address in machine translation fine-tuning?\",\"answer\":\"It targets improving machine translation by selecting more useful training data, since irrelevant or poorly translated pairs can degrade model performance. The paper also emphasizes making training more data-efficient.\"},{\"question\":\"How does the proposed method evaluate the usefulness of training data?\",\"answer\":\"It uses a learnability score derived from the learner model and leverages a pre-trained reference model. The approach scores each data point’s utility for fine-tuning.\"},{\"question\":\"What is the benefit of batch selection in the method?\",\"answer\":\"Batch selection explicitly considers interdependencies among data points, making it more effective than independently selecting top-scoring samples. The experiments show improved data efficiency and translation performance versus iid and random baselines.\"}]","Dynamic Jointly Batch Selection for Data Efficient Machine Translation Fine-Tuning | PDF",1791035471,23]