This study addresses the low-resource Indian lan- 002guage translation task (English Assamese, English Ma- 003nipuri) at WMT 2025, proposing a cross-iterative back- 004translation and data augmentation approach based on 005dual pre-trained models to enhance translation perfor- 006mance in low-resource scenarios. The research method- 007ology primarily encompasses four aspects: (1) Utilizing 008open-source pre-trained models IndicTrans2_1B and 009NLLB_3.3B, fine-tuning them on official bilingual data, 010followed by alternating back-translation and incremen- 011tal training to generate high-quality pseudo-parallel cor- 012pora and optimize model parameters through multiple 013iterations; (2) Employing the open-source semantic sim- 014ilarity model (all-mpnet-base-v2) to filter monolingual 015sentences with low semantic similarity to the test set 016from open-source corpora such as NLLB and BPCC, 017thereby improving the relevance of monolingual data 018to the task; (3) Cleaning the training data, including 019removing URL and HTML format content, eliminating 020untranslated sentences in back-translation, standardiz- 021ing symbol formats, and normalizing capitalization of 022the first letter; (4) During the model inference phase, 023combining the outputs generated by the fine-tuned In- 024dicTrans2_1B and NLLB3.3B

The rapid advancement of scientific progress requires innovative tools that can accelerate knowledge discovery. Although recent AI methods, particularly large language models (LLMs), have shown promise in tasks such as hypothesis generation and experimental design, they fall short of replicating the collaborative nature of real-world scientific practices, where diverse experts work together in teams to tackle complex problems. To address the limitations, we propose an LLM-based multi-agent system, i.e., Virtual Scientists (VIRSCI), designed to mimic the teamwork inherent in scientific research. VIRSCI organizes a team of agents to collaboratively generate, evaluate, and refine research ideas. Through comprehensive experiments, we demonstrate that this multi-agent approach outperforms the state-of-the-art method in producing novel scientific ideas. We further investigate the collaboration mechanisms that contribute to its tendency to produce ideas with higher novelty, offering valuable insights to guide future research and illuminating pathways toward building a robust system for autonomous scientific discovery. The code is available at https://github.com/open-sciencelab/Virtual-Scientists.