搜索结果：ability

Large language models can be continually pre-trained or fine-tuned to improve performance in specific domains, languages, or skills, but this specialization often degrades other capabilities and may cause catastrophic forgetting. We investigate how abilities are distributed within LLM parameters by analyzing module activations under domain- and language-specific inputs for closely related models. Across layers and modules, we find that ability-related activations are highly concentrated in a small set of channels (typically <5\%), and these channels are largely disentangled with good sufficiency and stability. Building on these observations, we propose ACT (Activation-Guided Channel-wise Ability Transfer), which localizes ability-relevant channels via activation differences and selectively transfers only the corresponding parameters, followed by lightweight fine-tuning for compatibility. Experiments on multilingual mathematical and scientific reasoning show that ACT can recover forgotten abilities while preserving retained skills. It can also merge multiple specialized models to integrate several abilities into a single model with minimal interference. Our code and data will be

This study proposes a segmental-level prosodic probing framework to evaluate neural TTS models' ability to reproduce consonant-induced f0 perturbation, a fine-grained segmental-prosodic effect that reflects local articulatory mechanisms. We compare synthetic and natural speech realizations for thousands of words, stratified by lexical frequency, using Tacotron 2 and FastSpeech 2 trained on the same speech corpus (LJ Speech). These controlled analyses are then complemented by a large-scale evaluation spanning multiple advanced TTS systems. Results show accurate reproduction for high-frequency words but poor generalization to low-frequency items, suggesting that the examined TTS architectures rely more on lexical-level memorization than on abstract segmental-prosodic encoding. This finding highlights a limitation in such TTS systems' ability to generalize prosodic detail beyond seen data. The proposed probe offers a linguistically informed diagnostic framework that may inform future TTS evaluation methods, and has implications for interpretability and authenticity assessment in synthetic speech.

Large language models (LLMs) have recently shown remarkable performance in language tasks and beyond. However, due to their limited inherent causal reasoning ability, LLMs still face challenges in handling tasks that require robust causal reasoning ability, such as health-care and economic analysis. As a result, a growing body of research has focused on enhancing the causal reasoning ability of LLMs. Despite the booming research, there lacks a survey to well review the challenges, progress and future directions in this area. To bridge this significant gap, we systematically review literature on how to strengthen LLMs' causal reasoning ability in this paper. We start from the introduction of background and motivations of this topic, followed by the summarisation of key challenges in this area. Thereafter, we propose a novel taxonomy to systematically categorise existing methods, together with detailed comparisons within and between classes of methods. Furthermore, we summarise existing benchmarks and evaluation metrics for assessing LLMs' causal reasoning ability. Finally, we outline future research directions for this emerging field, offering insights and inspiration to researchers

Programming ability is one of the most important abilities for the undergraduates majoring in computer science. Taking Yunnan University as an example, the necessity and importance of improving the ability of programming is analyzed in this paper. The exploration and practice of improving students' ability of programming are discussed from four aspects: arrangement and reform of programming curriculums, construction of online programming practice innovation platform, certification of programming ability and organization of programming competitions. These reforms have achieved good results in recent years, which can provide reference for the practical teaching reform of computer specialty in relevant universities.

As multimodal large language models (MLLMs) advance rapidly, rigorous evaluation has become essential, providing further guidance for their development. In this work, we focus on a unified and robust evaluation of \textbf{vision perception} abilities, the foundational skill of MLLMs. We find that existing perception benchmarks, each focusing on different question types, domains, and evaluation metrics, introduce significant evaluation variance, complicating comprehensive assessments of perception abilities when relying on any single benchmark. To address this, we introduce \textbf{AbilityLens}, a unified benchmark designed to evaluate MLLMs in six key perception abilities (ranging from counting, OCR, to understanding structural data), focusing on both accuracy and stability, with each ability encompassing diverse types of questions, domains, and metrics. With the assistance of AbilityLens, we: (1) identify the strengths and weaknesses of current main-stream MLLMs, highlighting stability patterns and revealing a notable performance gap between state-of-the-art open-source and closed-source models; (2) uncover interesting ability conflict and early convergence phenomena during MLLM t

Multi-lingual ability transfer has become increasingly important for the broad application of large language models (LLMs). Existing work highly relies on training with the multi-lingual ability-related data, which may not be available for low-resource languages. To solve it, we propose a Multi-lingual Abilities Extraction and Combination approach, named as MAEC. Our key idea is to decompose and extract language-agnostic ability-related weights from LLMs, and combine them across different languages by simple addition and subtraction operations without training. Specifically, our MAEC consists of the extraction and combination stages. In the extraction stage, we firstly locate key neurons that are highly related to specific abilities, and then employ them to extract the transferable ability-related weights. In the combination stage, we further select the ability-related tensors that mitigate the linguistic effects, and design a combining strategy based on them and the language-specific weights, to build the multi-lingual ability-enhanced LLM. To assess the effectiveness of our approach, we conduct extensive experiments on LLaMA-3 8B on mathematical and scientific tasks in both high-

In-Context Learning (ICL) is a critical capability of Large Language Models (LLMs) as it empowers them to comprehend and reason across interconnected inputs. Evaluating the ICL ability of LLMs can enhance their utilization and deepen our understanding of how this ability is acquired at the training stage. However, existing evaluation frameworks primarily focus on language abilities and knowledge, often overlooking the assessment of ICL ability. In this work, we introduce the ICLEval benchmark to evaluate the ICL abilities of LLMs, which encompasses two key sub-abilities: exact copying and rule learning. Through the ICLEval benchmark, we demonstrate that ICL ability is universally present in different LLMs, and model size is not the sole determinant of ICL efficacy. Surprisingly, we observe that ICL abilities, particularly copying, develop early in the pretraining process and stabilize afterward. Our source codes and benchmark are released at https://github.com/yiye3/ICLEval.

LLM-as-a-Judge leverages the generative and reasoning capabilities of large language models (LLMs) to evaluate LLM responses across diverse scenarios, providing accurate preference signals. This approach plays a vital role in aligning LLMs with human values, ensuring ethical and reliable AI outputs that align with societal norms. Recent studies have raised many methods to train LLM as generative judges, but most of them are data consuming or lack accuracy, and only focus on LLM's judge ability. In this work, we regard judge ability as a general ability of LLM and implement a two-stage training approach, comprising supervised fine-tuning (SFT) warm-up and direct preference optimization (DPO) enhancement, to achieve judge style adaptation and improve judgment accuracy. Additionally, we introduce an efficient data synthesis method to generate judgmental content. Experimental results demonstrate that our approach, utilizing only about 2% to 40% of the data required by other methods, achieves SOTA performance on RewardBench. Furthermore, our training method enhances the general capabilities of the model by constructing complicated judge task, and the judge signals provided by our model

Practical research was conducted to cultivate students' mathematical computation ability using literature analysis, theoretical practice, and statistical analysis methods. The research involved 171 ninth-grade students from the author's school and was divided into two experimental groups (A and B) and a control group (C). The study aimed to improve students' mathematical computation ability through algorithm analysis and comparison. The results showed that after a long period of practice, there was an improvement in students' computational ability, and most students could reach the level of high school mathematical computation ability. However, further research is needed to determine how to achieve level three of mathematical computation ability. The study found that students' computational ability levels were similar in the experimental and control groups at the beginning of the study. Through targeted discussion courses, students' computational ability levels were improved. The degree of improvement was not significantly related to students' gender but had a moderate positive correlation with the number of times students participated in related discussions. Based on the practical

Recent studies have shown that Large Language Models (LLMs) have the potential to process extremely long text. Many works only evaluate LLMs' long-text processing ability on the language modeling task, with perplexity (PPL) as the evaluation metric. However, in our study, we find that there is no correlation between PPL and LLMs' long-text understanding ability. Besides, PPL may only reflect the model's ability to model local information instead of catching long-range dependency. Therefore, only using PPL to prove the model could process long text is inappropriate. The local focus feature of PPL could also explain some existing phenomena, such as the great extrapolation ability of the position method ALiBi. When evaluating a model's ability in long text, we might pay more attention to PPL's limitation and avoid overly relying on it.

Multi-modal large language models (MLLMs) can understand image-language prompts and demonstrate impressive reasoning ability. In this paper, we extend MLLMs' output by empowering MLLMs with the segmentation ability. The extended MLLMs can both output language responses to the image-language prompts and segment the regions that the complex question or query in the language prompts focuses on. To this end, the existing work, LISA, enlarges the original word embeddings with an additional segment token and fine-tunes dialogue generation and query-focused segmentation together, where the feature of the segment token is used to prompt the segment-anything model. Although they achieve superior segmentation performance, we observe that the dialogue ability decreases by a large margin compared to the original MLLMs. To maintain the original MLLMs' dialogue ability, we propose a novel MLLMs framework, coined as LLaVASeg, which leverages a chain-of-thought prompting strategy to instruct the MLLMs to segment the target region queried by the user. The MLLMs are first prompted to reason about the simple description of the target region from the complicated user query, then extract the visual att

Auxiliary function is a helpful component to improve language model's code generation ability. However, a systematic exploration of how they affect has yet to be done. In this work, we comprehensively evaluate the ability to utilize auxiliary functions encoded in recent code-pretrained language models. First, we construct a human-crafted evaluation set, called HumanExtension, which contains examples of two functions where one function assists the other. With HumanExtension, we design several experiments to examine their ability in a multifaceted way. Our evaluation processes enable a comprehensive understanding of including auxiliary functions in the prompt in terms of effectiveness and robustness. An additional implementation style analysis captures the models' various implementation patterns when they access the auxiliary function. Through this analysis, we discover the models' promising ability to utilize auxiliary functions including their self-improving behavior by implementing the two functions step-by-step. However, our analysis also reveals the model's underutilized behavior to call the auxiliary function, suggesting the future direction to enhance their implementation by e

The first step in evaluating a potential diagnostic biomarker is to examine the variation in its values across different disease groups. In a three-class disease setting, the volume under the receiver operating characteristic surface and the three-class Youden index are commonly used summary measures of a biomarker's discriminatory ability. However, these measures rely on a stochastic ordering assumption for the distributions of biomarker outcomes across the three groups. This assumption can be restrictive, particularly when covariates are involved, and its violation may lead to incorrect conclusions about a biomarker's ability to distinguish between the three disease classes. Even when a stochastic ordering exists, the order may vary across different biomarkers in discovery studies involving dozens or even thousands of candidate biomarkers, complicating automated ranking. To address these challenges and complement existing measures, we propose the underlap coefficient, a novel summary index of a biomarker's ability to distinguish between three (or more) disease groups, and study its properties. Additionally, we introduce Bayesian nonparametric estimators for both the unconditional

Categorization, a core cognitive ability in humans that organizes objects based on common features, is essential to cognitive science as well as computer vision. To evaluate the categorization ability of visual AI models, various proxy tasks on recognition from datasets to open world scenarios have been proposed. Recent development of Large Multimodal Models (LMMs) has demonstrated impressive results in high-level visual tasks, such as visual question answering, video temporal reasoning, etc., utilizing the advanced architectures and large-scale multimodal instruction tuning. Previous researchers have developed holistic benchmarks to measure the high-level visual capability of LMMs, but there is still a lack of pure and in-depth quantitative evaluation of the most fundamental categorization ability. According to the research on human cognitive process, categorization can be seen as including two parts: category learning and category use. Inspired by this, we propose a novel, challenging, and efficient benchmark based on composite blocks, called ComBo, which provides a disentangled evaluation framework and covers the entire categorization process from learning to use. By analyzing t

With the development of science and the continuous progress of artificial intelligence technology, Large Language Models (LLMs) have begun to be widely utilized across various fields. However, in the field of psychological counseling, the ability of LLMs have not been systematically assessed. In this study, we assessed the psychological counseling ability of mainstream LLMs using 1096 psychological counseling skill questions which were selected from the Chinese National Counselor Level 3 Examination, including Knowledge-based, Analytical-based, and Application-based question types. The analysis showed that the correctness rates of the LLMs for Chinese questions, in descending order, were GLM-3 (46.5%), GPT-4 (46.1%), Gemini (45.0%), ERNIE-3.5 (45.7%) and GPT-3.5 (32.9%). The correctness rates of the LLMs for English questions, in descending order, were ERNIE-3.5 (43.9%), GPT-4 (40.6%), Gemini (36.6%), GLM-3 (29.9%) and GPT-3.5 (29.5%). A chi-square test indicated significant differences in the LLMs' performance on Chinese and English questions. Furthermore, we subsequently utilized the Counselor's Guidebook (Level 3) as a reference for ERNIE-3.5, resulting in a new correctness rate

Abstraction ability is crucial in human intelligence, which can also benefit various tasks in NLP study. Existing work shows that LLMs are deficient in abstract ability, and how to improve it remains unexplored. In this work, we design the framework AbsInstruct to enhance LLMs' abstraction ability through instruction tuning. The framework builds instructions with in-depth explanations to assist LLMs in capturing the underlying rationale of abstraction. Meanwhile, we introduce a plausibility estimator to select instructions that are more consistent with the abstraction knowledge of LLMs to be aligned. Then, our framework combines abstraction instructions with general-purpose ones to build a hybrid dataset. Extensive experiments and analyses demonstrate that our framework can considerably enhance LLMs' abstraction ability with strong generalization performance while maintaining their general instruction-following abilities.

Small Vision Language Models (SVLMs) generally refer to models with parameter sizes less than or equal to 2B. Their low cost and power consumption characteristics confer high commercial value. However, their reasoning abilities are limited by the number of parameters. To address this issue, this paper proposes a post-training optimization paradigm called the Incremental Training Strategy to enhance the reasoning ability of SVLMs. Firstly, we constructed a Self-Supervised Chain-of-Thought (COT) Data Construction System, which leverages multiple LVLMs with 7B parameters or more to transform original data into COT data in a self-supervised manner. Our proposed Incremental Training Strategy consists of four stages. Stage 1 injects domain knowledge by performing Supervised Fine-Tuning (SFT) to the pretrained model on the COT data. Stage 2 aligns the COT data format by conducting a small amount of Group Relative Policy Optimization (GRPO) training constrained only by format rewards on the COT data. Stage 3 enhances reasoning ability by applying GRPO training on the COT data with constraints on both format and accuracy rewards. The resulting model shows significant improvement compared to

Predicting the glass-forming ability (GFA) of chemical compositions remains a fundamental challenge in materials science, especially for oxide glasses with broad compositional diversity. Traditional empirical and thermodynamic approaches often fail to capture the complex, nonlinear factors governing vitrification. In this study, we applied two ensemble machine learning algorithms-Random Forest (RF) and Extreme Gradient Boosting (XGB)-to the glass_ternary_hipt dataset to predict the GFA of ternary oxide glasses directly from composition-derived descriptors. Both models achieved excellent predictive accuracy (R^2 > 0.92, MAE < 0.04), confirming that GFA is learnable from compositional features alone. Feature importance analysis revealed that electronegativity variance, atomic size mismatch, and valence electron descriptors are the most influential factors, while cohesive energy and ionic radius provided secondary contributions. These chemically interpretable features align with established theories of glass formation, thereby bridging predictive performance with physical understanding. The novelty of this work lies in systematically extending ML-based predictive modeling to ter

In this paper, the control ability with time attributy for the linear continuous-time (LCT) systems are defined and analyzed by the volume computing for the controllability region. Firstly, a relation theorem about the open-loop control ability, the control strategy space (\textit{i.e.}, the solution space of the input variable for control problems), and the some closed-loop performance for the LCT systems is purposed and proven. This theorem shows us the necessity to optimize the control ability for the practical engineering problems. Secondly, recurssive volume-computing algorithms with the low computing complexities for the finite-time controllability region are discussed. Finally, two analytical volume computations of the infinite-time controllability region for the systems with $n$ different and repeated real eigenvalues are deduced, and then by deconstructing the volume computing equations, 3 classes of the shape factors are constructed. These analytical volume and shape factors can describe accurately the size and shape of the controllability region. Because the time-attribute control ability for LCT systems is directly related to the controllability region with the unit inp