The rapid advancement of Large Language Models (LLMs) has established language as a core general-purpose cognitive substrate, driving the demand for specialized Language Processing Units (LPUs) tailored for LLM inference. To overcome the growing energy consumption of LLM inference systems, this paper proposes a Hardwired-Neurons Language Processing Unit (HNLPU), which physically hardwires LLM weight parameters into the computational fabric, achieving several orders of magnitude computational efficiency improvement by extreme specialization. However, a significant challenge still lies in the scale of modern LLMs. A straightforward hardwiring of GPT-OSS-120B would require fabricating photomask sets valued at over 6 billion dollars, rendering this straightforward solution economically impractical.

As a major disease threatening global public mental health, depression requires early detection and intervention to prevent the deterioration of conditions and even to save lives. However, the existing automated depression detection research using multimodal data faces multiple challenges, such as the small sample size of available public datasets and the challenge of effectively integrating different modalities. In this paper, we focus on multilingual depression detection and construct a multilingual dataset, M-DEA, based on several monolingual datasets. Based on the multilingual dataset, a multilingual depression detection framework integrating with audios and texts data is proposed, which utilizes Large Language Model for multilingual alignment. Additionally, a multimodal multi-space feature fusion method (MMS) is proposed. The method projects audio and text features into two feature sub-spaces and dynamically fuses features from different modalities and spaces through a multi-head attention mechanism, thereby improving the model's feature representation, capture, and fusion abilities. The method is able to learn the commonalities and differences of depression features in different languages, thus enabling effective multilingual depression detection. Meanwhile, by introducing datasets from other languages to assist model training, the method can further improve detection performance in monolingual scenarios.

Suicide has long been a severe public health problem worldwide. According to World Population Review, China accounts for over one-quarter of suicides worldwide, making it the fifth leading cause of death [1] , and calling for more attention to suicide detection and intervention research.

Automatic code generation plays an important role in ensuring the reliability and correctness of software programs. Reliable programs can be obtained automatically from verified program specifications by code generators. The target languages of the existing code generators are mainly functional languages, which are relatively less used than C/C[Formula: see text]. As C/C[Formula: see text] is widely used in the industry and many fundamental software facilities and the correctness verification of C/C[Formula: see text] programs is difficult and cumbersome, this paper provides an automatic conversion framework that allows to generate C[Formula: see text] implementation from verified Isabelle/HOL specifications. The framework is characterized by combining the verification convenience of Isabelle/HOL and the efficiency of C[Formula: see text]. Since the correctness of the functional Isabelle/HOL specification can be guaranteed by interactive proofs, the correctness of the relevant generated C[Formula: see text] implementation can also be maintained.

Knowledge graphs usually contain much implicit semantic information, which need to be further recognized through semantic inference. However, existing approaches are either not good at processing large scale data or not powerful enough for digging hidden relations thoroughly. This paper proposes a distributed OWL2 RL/RDF rule-based theory closure reasoning algorithm, named KGRL, for recognizing hidden relations in knowledge graphs. Since hidden relations derived from knowledge graph usually contain a lot of redundancies, a redundancy reduction strategy is proposed for eliminating redundant data without effect further queries on the knowledge graph. Extensive experiments and comprehensive evaluations are conducted. The experimental result shows that KGRL recognizes more hidden relations efficiently than Cichlid at different scales of the LUBM benchmark, and it only has a constant increase of runtime. Further more, the redundancy reduction strategy effectively reduces the size of the resulting knowledge graphs of hidden relation recognition on both synthetic and real-world knowledge graphs.