arXiv subjects

Xiaobing Zhou

Publications and source records attributed to Xiaobing Zhou.

3 recordsLinked to original sources

arXiv2026

Comparing Ocean Forecasts Driven with Machine Learning-based and Physics-based Atmospheric Forcings

Operational ocean forecasting systems conventionally employ dynamical ocean models driven by atmospheric forcing derived from numerical weather prediction (NWP) models. Recent advancements in artificial intelligence and machine learning (ML) have led to the development of ML-based atmospheric weather models, which have competitive, if not better, medium range forecast accuracy compared to traditional NWP systems. This study evaluates the impact of ML-based atmospheric forcing on ocean forecast skill through two sets of 10-day forecasts using the UK Met Office GOSI9 configuration of the NEMO dynamical ocean model. Both experiments share identical ocean initial conditions; but differ in atmospheric forcing: one uses ECMWF's ML-based AIFS model, while the other uses the Australian Bureau of Meteorology's physics-based NWP model, ACCESS-G3. Forecasts were initialized on the first day of each month over the period 2023-2024. The quality of the atmospheric forcing was assessed by comparing AIFS and ACCESS-G3 forecast skill against both ECMWF reanalysis v5 (ERA5) and ACCESS-G3 analyses. Results indicate that AIFS consistently outperforms ACCESS-G3, either from the initial forecast time or after the first few days. Oceanic forecast skill was evaluated against both the GOSI9 reanalysis and observations, focusing on key surface variables including sea surface temperature, salinity, sea level, and ocean currents. The ocean forecasts forced with AIFS atmospheric data exhibit comparable or enhanced predictive skill compared to those forced with ACCESS-G3 data. These findings underscore the potential of ML-based atmospheric models to replace traditional NWP forcing in operational ocean forecasting systems, offering improved accuracy and computational efficiency.

physics.ao-ph↗

arXiv2026

SAPO: Self-Adaptive Process Optimization Makes Small Reasoners Stronger

Existing self-evolution methods overlook the influence of fine-grained reasoning steps, which leads to the reasoner-verifier gap. The computational inefficiency of Monte Carlo (MC) process supervision further exacerbates the difficulty in mitigating the gap. Motivated by the Error-Related Negativity (ERN), which the reasoner can localize error following incorrect decisions, guiding rapid adjustments, we propose a Self-Adaptive Process Optimization (SAPO) method for self-improvement in Small Language Models (SLMs). SAPO adaptively and efficiently introduces process supervision signals by actively minimizing the reasoner-verifier gap rather than relying on inefficient MC estimations. Extensive experiments demonstrate that the proposed method outperforms most existing self-evolution methods on two challenging task types: mathematics and code. Additionally, to further investigate SAPO's impact on verifier performance, this work introduces two new benchmarks for process reward models in both mathematical and coding tasks.

cs.CL↗

arXiv2024

Enhancing Semantics in Multimodal Chain of Thought via Soft Negative Sampling

Chain of thought (CoT) has proven useful for problems requiring complex reasoning. Many of these problems are both textual and multimodal. Given the inputs in different modalities, a model generates a rationale and then uses it to answer a question. Because of the hallucination issue, the generated soft negative rationales with high textual quality but illogical semantics do not always help improve answer accuracy. This study proposes a rationale generation method using soft negative sampling (SNSE-CoT) to mitigate hallucinations in multimodal CoT. Five methods were applied to generate soft negative samples that shared highly similar text but had different semantics from the original. Bidirectional margin loss (BML) was applied to introduce them into the traditional contrastive learning framework that involves only positive and negative samples. Extensive experiments on the ScienceQA dataset demonstrated the effectiveness of the proposed method. Code and data are released at https://github.com/zgMin/SNSE-CoT.

cs.CL↗