Reassessing Layer Pruning in LLMs: New Insights and Methods

Although large language models (LLMs) have achieved remarkable success across various domains, their considerable scale necessitates substantial computational resources, posing significant challenges for deployment in resource-constrained environments. Layer pruning, as a simple yet effective compression method, removes layers of a model directly, reducing computational overhead. However, what are the best practices for layer pruning in LLMs? Are sophisticated layer selection metrics truly effective? Does the LoRA (Low-Rank Approximation) family, widely regarded as a leading method for pruned model fine-tuning, truly meet expectations when applied to post-pruning fine-tuning? To answer these questions, we dedicate thousands of GPU hours to benchmarking layer pruning in LLMs and gaining insights across multiple dimensions. Our results demonstrate that a simple approach, i.e., pruning the final 25\% of layers followed by fine-tuning the \texttt{lm\_head} and the remaining last three layer, yields remarkably strong performance. Following this guide, we prune Llama-3.1-8B-It and obtain a model that outperforms many popular LLMs of similar size, such as ChatGLM2-6B, Vicuna-7B-v1.5, Qwen1.5-7B and Baichuan2-7B. We release the optimal model weights on Huggingface, and the code is available on GitHub.

MDM: Advancing Multi-Domain Distribution Matching for Automatic Modulation Recognition Dataset Synthesis

Recently, deep learning technology has been successfully introduced into Automatic Modulation Recognition (AMR) tasks. However, the success of deep learning is all attributed to the training on large-scale datasets. Such a large amount of data brings huge pressure on storage, transmission and model training. In order to solve the problem of large amount of data, some researchers put forward the method of data distillation, which aims to compress large training data into smaller synthetic datasets to maintain its performance. While numerous data distillation techniques have been developed within the realm of image processing, the unique characteristics of signals set them apart. Signals exhibit distinct features across various domains, necessitating specialized approaches for their analysis and processing. To this end, a novel dataset distillation method--Multi-domain Distribution Matching (MDM) is proposed. MDM employs the Discrete Fourier Transform (DFT) to translate timedomain signals into the frequency domain, and then uses a model to compute distribution matching losses between the synthetic and real datasets, considering both the time and frequency domains. Ultimately, these two losses are integrated to update the synthetic dataset. We conduct extensive experiments on three AMR datasets. Experimental results show that, compared with baseline methods, our method achieves better performance under the same compression ratio. Furthermore, we conduct crossarchitecture generalization experiments on several models, and the experimental results show that our synthetic datasets can generalize well on other unseen models.

A Generic Layer Pruning Method for Signal Modulation Recognition Deep Learning Models

With the successful application of deep learning in communications systems, deep neural networks are becoming the preferred method for signal classification. Although these models yield impressive results, they often come with high computational complexity and large model sizes, which hinders their practical deployment in communication systems. To address this challenge, we propose a novel layer pruning method. Specifically, we decompose the model into several consecutive blocks, each containing consecutive layers with similar semantics. Then, we identify layers that need to be preserved within each block based on their contribution. Finally, we reassemble the pruned blocks and fine-tune the compact model. Extensive experiments on five datasets demonstrate the efficiency and effectiveness of our method over a variety of state-of-the-art baselines, including layer pruning and channel pruning methods.

Mixing Signals: Data Augmentation Approach for Deep Learning Based Modulation Recognition

With the rapid development of deep learning, automatic modulation recognition (AMR), as an important task in cognitive radio, has gradually transformed from traditional feature extraction and classification to automatic classification by deep learning technology. However, deep learning models are data-driven methods, which often require a large amount of data as the training support. Data augmentation, as the strategy of expanding dataset, can improve the generalization of the deep learning models and thus improve the accuracy of the models to a certain extent. In this paper, for AMR of radio signals, we propose a data augmentation strategy based on mixing signals and consider four specific methods (Random Mixing, Maximum-Similarity-Mixing, $\theta-$Similarity Mixing and n-times Random Mixing) to achieve data augmentation. Experiments show that our proposed method can improve the classification accuracy of deep learning based AMR models in the full public dataset RML2016.10a. In particular, for the case of a single signal-to-noise ratio signal set, the classification accuracy can be significantly improved, which verifies the effectiveness of the methods.

Deep Transfer Clustering of Radio Signals

Modulation recognition is an important task in radio signal processing. Most of the current researches focus on supervised learning. However, in many real scenarios, it is difficult and cost to obtain the labels of signals. In this letter, we turn to the more challenging problem: can we cluster the modulation types just based on a large number of unlabeled radio signals? If this problem can be solved, we then can also recognize modulation types by manually labeling a very small number of samples. To answer this problem, we propose a deep transfer clustering (DTC) model. DTC naturally integrates feature learning and deep clustering, and further adopts a transfer learning mechanism to improve the feature extraction ability of an embedded convolutional neural network (CNN) model. The experiments validate that our DTC significantly outperforms a number of baselines, achieving the state-of-the-art performance in clustering radio signals for modulation recognition.

Adaptive Visibility Graph Neural Network and its Application in Modulation Classification

Our digital world is full of time series and graphs which capture the various aspects of many complex systems. Traditionally, there are respective methods in processing these two different types of data, e.g., Recurrent Neural Network (RNN) and Graph Neural Network (GNN), while in recent years, time series could be mapped to graphs by using the techniques such as Visibility Graph (VG), so that researchers can use graph algorithms to mine the knowledge in time series. Such mapping methods establish a bridge between time series and graphs, and have high potential to facilitate the analysis of various real-world time series. However, the VG method and its variants are just based on fixed rules and thus lack of flexibility, largely limiting their application in reality. In this paper, we propose an Adaptive Visibility Graph (AVG) algorithm that can adaptively map time series into graphs, based on which we further establish an end-to-end classification framework AVGNet, by utilizing GNN model DiffPool as the classifier. We then adopt AVGNet for radio signal modulation classification which is an important task in the field of wireless communication. The simulations validate that AVGNet outperforms a series of advanced deep learning methods, achieving the state-of-the-art performance in this task.