DeepSeek, a Chinese artificial intelligence (AI) startup, has released their V3 and R1 series models, which attracted global attention due to their low cost, high performance, and open-source advantages. This paper begins by reviewing the evolution of large AI models focusing on paradigm shifts, the mainstream large language model (LLM) paradigm, and the DeepSeek paradigm. Subsequently, the paper highlights novel algorithms introduced by DeepSeek, including multi-head latent attention (MLA), mixture-of-experts (MoE), multi-token prediction (MTP), and group relative policy optimization (GRPO). The paper then explores DeepSeek's engineering breakthroughs in LLM scaling, training, inference, and system-level optimization architecture. Moreover, the impact of DeepSeek models on the competitive AI landscape is analyzed, comparing them to mainstream LLMs across various fields. Finally, the paper reflects on the insights gained from DeepSeek's innovations and discusses future trends in the technical and engineering development of large AI models, particularly in data, training, and reasoning.

Data-to-text Generation (D2T) aims to generate textual natural language statements that can fluently and precisely describe the structured data such as graphs, tables, and meaning representations (MRs) in the form of key-value pairs. It is a typical and crucial task in natural language generation (NLG). Early D2T systems generated texts with the cost of human engineering in designing domain specific rules and templates, and achieved acceptable performance in coherence, fluency, and fidelity. In recent years, the data-driven D2T systems based on deep learning have reached state-of-the-art (SOTA) performance in more challenging datasets. In this paper, we provide a comprehensive review on existing neural data-to-text generation approaches. We first introduce available D2T resources, including systematically categorized D2T datasets and mainstream evaluation metrics. Next, we survey existing works based on the taxonomy along two axes: neural end-to-end D2T and neural modular D2T. We also discuss the potential applications and the adverse impacts. Finally, we present readers with the challenges faced by neural D2T and outline some potential future directions in this area.

Thanks to the development of geographic information technology, geospatial representation learning based on POIs (Point-of-Interest) has gained widespread attention in the past few years. POI is an important indicator to reflect urban socioeconomic activities, widely used to extract geospatial information. However, previous studies often focus on a specific area, such as a city or a district, and are designed only for particular tasks, such as land-use classification. On the other hand, large-scale pre-trained models (PTMs) have recently achieved impressive success and become a milestone in artificial intelligence (AI). Against this background, this study proposes the first large-scale pre-training geospatial representation learning model called GeoBERT. First, we collect about 17 million POIs in 30 cities across China to construct pre-training corpora, with 313 POI types as the tokens and the level-7 Geohash grids as the basic units. Second, we pre-train GeoEBRT to learn grid embedding in self-supervised learning by masking the POI type and then predicting. Third, under the paradigm of “pre-training + fine-tuning”, we design five practical downstream tasks. Experiments show that, with just one additional output layer fine-tuning, GeoBERT outperforms previous NLP methods (Word2vec, GloVe) used in geospatial representation learning by 9.21% on average in F1-score for classification tasks, such as store site recommendation and working/living area prediction. For regression tasks, such as POI number prediction, house price prediction, and passenger flow prediction, GeoBERT demonstrates greater performance improvements. The experiment results prove that pre-training on large-scale POI data can significantly improve the ability to extract geospatial information. In the discussion section, we provide a detailed analysis of what GeoBERT has learned from the perspective of attention mechanisms.

Commit messages are natural language descriptions of code changes, which are important for software evolution such as code understanding and maintenance. However, previous methods are trained on the entire dataset without considering the fact that a portion of commit messages adhere to good practice (i.e., good-practice commits), while the rest do not. On the basis of our empirical study, we discover that training on good-practice commits significantly contributes to the commit message generation. Motivated by this finding, we propose a novel knowledge-aware denoising learning method called KADEL. Considering that good-practice commits constitute only a small proportion of the dataset, we align the remaining training samples with these good-practice commits. To achieve this, we propose a model that learns the commit knowledge by training on good-practice commits. This knowledge model enables supplementing more information for training samples that do not conform to good practice. However, since the supplementary information may contain noise or prediction errors, we propose a dynamic denoising training method. This method composes a distribution-aware confidence function and a dynamic distribution list, which enhances the effectiveness of the training process. Experimental results on the whole MCMD dataset demonstrate that our method overall achieves state-of-the-art performance compared with previous methods.

Knowledge Representation Learning (KRL) is crucial for enabling applications of symbolic knowledge from Knowledge Graphs (KGs) to downstream tasks by projecting knowledge facts into vector spaces. Despite their effectiveness in modeling KG structural information, KRL methods are suffering from the sparseness of KGs. The rise of Large Language Models (LLMs) built on the Transformer architecture presents promising opportunities for enhancing KRL by incorporating textual information to address information sparsity in KGs. LLM-enhanced KRL methods, including three key approaches, encoder-based methods that leverage detailed contextual information, encoder-decoder-based methods that utilize a unified Seq2Seq model for comprehensive encoding and decoding, and decoder-based methods that utilize extensive knowledge from large corpora, have significantly advanced the effectiveness and generalization of KRL in addressing a wide range of downstream tasks. This work provides a broad overview of downstream tasks while simultaneously identifying emerging research directions in these evolving domains.