This study proposes a method for distilling the knowledge of fine-tuned Large Language Models (LLMs) into a smaller, more efficient, and accurate neural network, specifically targeting the challenge of deploying these models on resource-constrained devices. Our methodology involves training the smaller student model using the prediction probabilities of the LLM, which serves as a teacher model. This is achieved through a specialized loss function tailored to learn from the LLM's output probabilities, ensuring that the student model closely mimics the teacher's performance. To test this approach, we utilized a large dataset, 7T, containing 6,684 student-written responses to science questions and three other datasets with student-written responses. We also compared performance with original neural network (NN) models to validate the accuracy. Results have shown that the NN and distilled student models have comparable accuracy to the teacher model for the 7T dataset; however, other datasets have shown significantly lower accuracy (28% on average) for NN, though our proposed distilled model is still able to achieve 12\% higher accuracy than NN. Furthermore, the student model size ranges from 0.1M to 0.02M, 100 times smaller in terms of parameters and ten times smaller compared with the original output model size. The significance of this research lies in its potential to make advanced AI technologies accessible in typical educational settings, particularly for automatic scoring.
相關內容
大語言模型是基于海量文本數據訓練的深度學習模型。它不僅能夠生成自然語言文本，還能夠深入理解文本含義，處理各種自然語言任務，如文本摘要、問答、翻譯等。2023年，大語言模型及其在人工智能領域的應用已成為全球科技研究的熱點，其在規模上的增長尤為引人注目，參數量已從最初的十幾億躍升到如今的一萬億。參數量的提升使得模型能夠更加精細地捕捉人類語言微妙之處，更加深入地理解人類語言的復雜性。在過去的一年里，大語言模型在吸納新知識、分解復雜任務以及圖文對齊等多方面都有顯著提升。隨著技術的不斷成熟，它將不斷拓展其應用范圍，為人類提供更加智能化和個性化的服務，進一步改善人們的生活和生產方式。
This study is based on the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset and aims to explore early detection and disease progression in Alzheimer's disease (AD). We employ innovative data preprocessing strategies, including the use of the random forest algorithm to fill missing data and the handling of outliers and invalid data, thereby fully mining and utilizing these limited data resources. Through Spearman correlation coefficient analysis, we identify some features strongly correlated with AD diagnosis. We build and test three machine learning models using these features: random forest, XGBoost, and support vector machine (SVM). Among them, the XGBoost model performs the best in terms of diagnostic performance, achieving an accuracy of 91%. Overall, this study successfully overcomes the challenge of missing data and provides valuable insights into early detection of Alzheimer's disease, demonstrating its unique research value and practical significance.
We introduce NeuRes, a neuro-symbolic proof-based SAT solver. Unlike other neural SAT solving methods, NeuRes is capable of proving unsatisfiability as opposed to merely predicting it. By design, NeuRes operates in a certificate-driven fashion by employing propositional resolution to prove unsatisfiability and to accelerate the process of finding satisfying truth assignments in case of unsat and sat formulas, respectively. To realize this, we propose a novel architecture that adapts elements from Graph Neural Networks and Pointer Networks to autoregressively select pairs of nodes from a dynamic graph structure, which is essential to the generation of resolution proofs. Our model is trained and evaluated on a dataset of teacher proofs and truth assignments that we compiled with the same random formula distribution used by NeuroSAT. In our experiments, we show that NeuRes solves more test formulas than NeuroSAT by a rather wide margin on different distributions while being much more data-efficient. Furthermore, we show that NeuRes is capable of largely shortening teacher proofs by notable proportions. We use this feature to devise a bootstrapped training procedure that manages to reduce a dataset of proofs generated by an advanced solver by ~23% after training on it with no extra guidance.
Mixture of Experts (MoE) models have emerged as a primary solution for reducing the computational cost of Large Language Models. In this work, we analyze their scaling properties, incorporating an expanded range of variables. Specifically, we introduce a new hyperparameter, granularity, whose adjustment enables precise control over the size of the experts. Building on this, we establish scaling laws for fine-grained MoE, taking into account the number of training tokens, model size, and granularity. Leveraging these laws, we derive the optimal training configuration for a given computational budget. Our findings not only show that MoE models consistently outperform dense Transformers but also highlight that the efficiency gap between dense and MoE models widens as we scale up the model size and training budget. Furthermore, we demonstrate that the common practice of setting the size of experts in MoE to mirror the feed-forward layer is not optimal at almost any computational budget.
We study the maximum likelihood (ML) degree of discrete exponential independence models and models defined by the second hypersimplex. For models with two independent variables, we show that the ML degree is an invariant of a matroid associated to the model. We use this description to explore ML degrees via hyperplane arrangements. For independence models with more variables, we investigate the connection between the vanishing of factors of its principal $A$-determinant and its ML degree. Similarly, for models defined by the second hypersimplex, we determine its principal $A$-determinant and give computational evidence towards a conjectured lower bound of its ML degree.
Arora & Ge introduced a noise-free polynomial system to compute the secret of a Learning With Errors (LWE) instance via linearization. Albrecht et al. later utilized the Arora-Ge polynomial model to study the complexity of Gr\"obner basis computations on LWE polynomial systems under the assumption of semi-regularity. In this paper we revisit the Arora-Ge polynomial and prove that it satisfies a genericity condition recently introduced by Caminata & Gorla, called being in generic coordinates. For polynomial systems in generic coordinates one can always estimate the complexity of DRL Gr\"obner basis computations in terms of the Castelnuovo-Mumford regularity and henceforth also via the Macaulay bound. Moreover, we generalize the Gr\"obner basis algorithm of Semaev & Tenti to arbitrary polynomial systems with a finite degree of regularity. In particular, existence of this algorithm yields another approach to estimate the complexity of DRL Gr\"obner basis computations in terms of the degree of regularity. In practice, the degree of regularity of LWE polynomial systems is not known, though one can always estimate the lowest achievable degree of regularity. Consequently, from a designer's worst case perspective this approach yields sub-exponential complexity estimates for general, binary secret, and binary error LWE. In recent works by Dachman-Soled et al. the hardness of LWE in the presence of side information was analyzed. Utilizing their framework we discuss how hints can be incorporated into LWE polynomial systems and how they affect the complexity of Gr\"obner basis computations.
This position paper investigates the integration of Differential Privacy (DP) in the training of Mixture of Experts (MoE) models within the field of natural language processing. As Large Language Models (LLMs) scale to billions of parameters, leveraging expansive datasets, they exhibit enhanced linguistic capabilities and emergent abilities. However, this growth raises significant computational and privacy concerns. Our study addresses these issues by exploring the potential of MoE models, known for their computational efficiency, and the application of DP, a standard for privacy preservation. We present the first known attempt to train MoE models under the constraints of DP, addressing the unique challenges posed by their architecture and the complexities of DP integration. Our initial experimental studies demonstrate that MoE models can be effectively trained with DP, achieving performance that is competitive with their non-private counterparts. This initial study aims to provide valuable insights and ignite further research in the domain of privacy-preserving MoE models, softly laying the groundwork for prospective developments in this evolving field.
This study proposes a method for knowledge distillation (KD) of fine-tuned Large Language Models (LLMs) into smaller, more efficient, and accurate neural networks. We specifically target the challenge of deploying these models on resource-constrained devices. Our methodology involves training the smaller student model (Neural Network) using the prediction probabilities (as soft labels) of the LLM, which serves as a teacher model. This is achieved through a specialized loss function tailored to learn from the LLM's output probabilities, ensuring that the student model closely mimics the teacher's performance. To validate the performance of the KD approach, we utilized a large dataset, 7T, containing 6,684 student-written responses to science questions and three mathematical reasoning datasets with student-written responses graded by human experts. We compared accuracy with state-of-the-art (SOTA) distilled models, TinyBERT, and artificial neural network (ANN) models. Results have shown that the KD approach has 1% and 4% higher scoring accuracy than ANN and TinyBERT and comparable accuracy to the teacher model. Furthermore, the student model size is 0.02M, 10,000 times smaller in parameters and x10 faster in inferencing than the teacher model and TinyBERT, respectively. The significance of this research lies in its potential to make advanced AI technologies accessible in typical educational settings, particularly for automatic scoring.
Emotion recognition in conversation (ERC) aims to detect the emotion label for each utterance. Motivated by recent studies which have proven that feeding training examples in a meaningful order rather than considering them randomly can boost the performance of models, we propose an ERC-oriented hybrid curriculum learning framework. Our framework consists of two curricula: (1) conversation-level curriculum (CC); and (2) utterance-level curriculum (UC). In CC, we construct a difficulty measurer based on "emotion shift" frequency within a conversation, then the conversations are scheduled in an "easy to hard" schema according to the difficulty score returned by the difficulty measurer. For UC, it is implemented from an emotion-similarity perspective, which progressively strengthens the model's ability in identifying the confusing emotions. With the proposed model-agnostic hybrid curriculum learning strategy, we observe significant performance boosts over a wide range of existing ERC models and we are able to achieve new state-of-the-art results on four public ERC datasets.
Reasoning with knowledge expressed in natural language and Knowledge Bases (KBs) is a major challenge for Artificial Intelligence, with applications in machine reading, dialogue, and question answering. General neural architectures that jointly learn representations and transformations of text are very data-inefficient, and it is hard to analyse their reasoning process. These issues are addressed by end-to-end differentiable reasoning systems such as Neural Theorem Provers (NTPs), although they can only be used with small-scale symbolic KBs. In this paper we first propose Greedy NTPs (GNTPs), an extension to NTPs addressing their complexity and scalability limitations, thus making them applicable to real-world datasets. This result is achieved by dynamically constructing the computation graph of NTPs and including only the most promising proof paths during inference, thus obtaining orders of magnitude more efficient models. Then, we propose a novel approach for jointly reasoning over KBs and textual mentions, by embedding logic facts and natural language sentences in a shared embedding space. We show that GNTPs perform on par with NTPs at a fraction of their cost while achieving competitive link prediction results on large datasets, providing explanations for predictions, and inducing interpretable models. Source code, datasets, and supplementary material are available online at //github.com/uclnlp/gntp.
Deep Convolutional Neural Networks have pushed the state-of-the art for semantic segmentation provided that a large amount of images together with pixel-wise annotations is available. Data collection is expensive and a solution to alleviate it is to use transfer learning. This reduces the amount of annotated data required for the network training but it does not get rid of this heavy processing step. We propose a method of transfer learning without annotations on the target task for datasets with redundant content and distinct pixel distributions. Our method takes advantage of the approximate content alignment of the images between two datasets when the approximation error prevents the reuse of annotation from one dataset to another. Given the annotations for only one dataset, we train a first network in a supervised manner. This network autonomously learns to generate deep data representations relevant to the semantic segmentation. Then the images in the new dataset, we train a new network to generate a deep data representation that matches the one from the first network on the previous dataset. The training consists in a regression between feature maps and does not require any annotations on the new dataset. We show that this method reaches performances similar to a classic transfer learning on the PASCAL VOC dataset with synthetic transformations.
小貼士
登錄享
相關主題
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191