Yang Song

Large Language Models (LLMs) have demonstrated strong potential in mathematical reasoning tasks. However, the performance remains highly sensitive to surface-level problem formulations, and the model training is also limited by sparse and coarse-grained supervision signals. In this paper, we propose an active reinforcement based low-resource reasoning approach (ARLRA), which is a closed-loop and data-efficient LLMs training framework for mathematical reasoning improvement. Specifically, a web based guiding expansion module firstly expands reliable supervision coverage while preserving mathematical correctness. Then, a scene-aware assignment module diagnoses model performance in step-wise symbolic reasoning and adaptively routes to appropriate enhancement strategy. Finally, a self-adaptive reward mechanism provides fine-grained and multi-dimensional feedback, where different reward dimensions continuously adjust training process to focus on the model capability deficits. Experiments on four standard benchmarks show that the proposed ARLRA outperforms strong low-resource baselines and approaches the performance of methods trained with significantly larger datasets.

The brain semantic decoding has received a surge of attention in the computer vision and neuroscience disciplines. However, existing techniques ignore the sparse and implicit semantic analysis issue of the brain signal, leading to coarse-grained brain semantic decoding. In this paper, visual guided dual-spatial interaction network (VDIN) is first proposed to significantly facilitate the fine-grained brain semantic decoding, with the multi-way guidance of the visual message (text and image). Specifically, the local dual-spatial interaction operation is proposed to exploit the explicit-implicit coupling semantic cues between brain and text, via the text-aware space. Moreover, the above operation simultaneously attends to the multi-level semantic cues between brain and image from the more high-resolution imagea-ware space. Furthermore, the global dual-spatial interaction operation is utilized to integrate and modify the obtained local semantic cues into the global one. This naturally boosts the investigation of visual-brain consistency and complementary, leading to the more expressive and fine-grained brain semantic decoding architecture. For evaluation, VDIN was verified on the public brain-visual benchmark EEGCVPR40, the experiments demonstrated that the proposed model outperforms the best baseline with an improvement of 15.97%. Note that, our model still achieves competitive performance even considering the brain spatial space associated with fewer channels.

Background: Long non-coding RNAs (lncRNAs) play an important role in angiogenesis, immune response, inflammatory response and tumor development and metastasis. m6 A (N6—methyladenosine) is one of the most common RNA modifications in eukaryotes. The aim of our research was to investigate the potential prognostic value of m6A-related lncRNAs in ovarian cancer (OC). Methods: The data we need for our research was downloaded from the Cancer Genome Atlas (TCGA) and the Gene Expression Omnibus (GEO) database. Pearson correlation analysis between 21 m6A regulators and lncRNAs was performed to identify m6A-related lncRNAs. Univariate Cox regression analysis was implemented to screen for lncRNAs with prognostic value. A least absolute shrinkage and selection operator (LASSO) Cox regression and multivariate Cox regression analyses was used to further reduct the lncRNAs with prognostic value and construct a m6A-related lncRNAs signature for predicting the prognosis of OC patients. Results: Two hundred seventy-five m6A-related lncRNAs were obtained using pearson correlation analysis. 29 m6A-related lncRNAs with prognostic value was selected through univariate Cox regression analysis. Then, a seven m6A-related lncRNAs signature was identified by LASSO Cox regression. Each patient obtained a riskscore through multivariate Cox regression analyses and the patients were classified into high-and low-risk group using the median riskscore as a cutoff. Kaplan–Meier curve revealed that the patients in high-risk group have poor outcome. The receiver operating characteristic curve revealed that the predictive potential of the m6A-related lncRNAs signature for OC was powerful. The predictive potential of the m6A-related lncRNAs signature was successfully validated in the GSE9891, GSE26193 datasets and our clinical specimens. Multivariate analyses suggested that the m6A-related lncRNAs signature was an independent prognostic factor for OC patients. Moreover, a nomogram based on the expression level of the seven m6A-related lncRNAs was established to predict survival rate of patients with OC. Finally, a competing endogenous RNA (ceRNA) network associated with the seven m6A-related lncRNAs was constructed to understand the possible mechanisms of the m6A-related lncRNAs involed in the progression of OC. Conclusions: In conclusion, our research revealed that the m6A-related lncRNAs may affect the prognosis of OC patients and identified a seven m6A-related lncRNAs signature to predict the prognosis of OC patients.