The upscaling of Large Language Models (LLMs) has yielded impressive advances in natural language processing, yet it also poses significant deployment challenges. Weight quantization has emerged as a widely embraced solution to reduce memory and computational demands. This paper introduces BitDistiller, a framework that synergizes Quantization-Aware Training (QAT) with Knowledge Distillation (KD) to boost the performance of LLMs at ultra-low precisions (sub-4-bit). Specifically, BitDistiller first incorporates a tailored asymmetric quantization and clipping technique to maximally preserve the fidelity of quantized weights, and then proposes a novel Confidence-Aware Kullback-Leibler Divergence (CAKLD) objective, which is employed in a self-distillation manner to enable faster convergence and superior model performance. Empirical evaluations demonstrate that BitDistiller significantly surpasses existing methods in both 3-bit and 2-bit configurations on general language understanding and complex reasoning benchmarks. Notably, BitDistiller is shown to be more cost-effective, demanding fewer data and training resources. The code is available at https://github.com/DD-DuDa/BitDistiller.
The incremental sequence labeling task involves continuously learning new classes over time while retaining knowledge of the previous ones. Our investigation identifies two significant semantic shifts: E2O (where the model mislabels an old entity as a non-entity) and O2E (where the model labels a non-entity or old entity as a new entity). Previous research has predominantly focused on addressing the E2O problem, neglecting the O2E issue. This negligence results in a model bias towards classifying new data samples as belonging to the new class during the learning process. To address these challenges, we propose a novel framework, Incremental Sequential Labeling without Semantic Shifts (IS3). Motivated by the identified semantic shifts (E2O and O2E), IS3 aims to mitigate catastrophic forgetting in models. As for the E2O problem, we use knowledge distillation to maintain the model's discriminative ability for old entities. Simultaneously, to tackle the O2E problem, we alleviate the model's bias towards new entities through debiased loss and optimization levels. Our experimental evaluation, conducted on three datasets with various incremental settings, demonstrates the superior performance of IS3 compared to the previous state-of-the-art method by a significant margin.The data, code, and scripts are publicly available at https://github.com/zzz47zzz/codebase-for-incremental-learning-with-llm.
This paper addresses the challenge of mitigating data heterogeneity among clients within a Federated Learning (FL) framework. The model-drift issue, arising from the noniid nature of client data, often results in suboptimal personalization of a global model compared to locally trained models for each client. To tackle this challenge, we propose a novel approach named FedD2S for Personalized Federated Learning (pFL), leveraging knowledge distillation. FedD2S incorporates a deep-to-shallow layer-dropping mechanism in the data-free knowledge distillation process to enhance local model personalization. Through extensive simulations on diverse image datasets-FEMNIST, CIFAR10, CINIC0, and CIFAR100-we compare FedD2S with state-of-the-art FL baselines. The proposed approach demonstrates superior performance, characterized by accelerated convergence and improved fairness among clients. The introduced layer-dropping technique effectively captures personalized knowledge, resulting in enhanced performance compared to alternative FL models. Moreover, we investigate the impact of key hyperparameters, such as the participation ratio and layer-dropping rate, providing valuable insights into the optimal configuration for FedD2S. The findings demonstrate the efficacy of adaptive layer-dropping in the knowledge distillation process to achieve enhanced personalization and performance across diverse datasets and tasks.
As the cornerstone of human-behavior analysis in video understanding, temporal action proposal generation aims to predict the starting and ending time of human action instances in untrimmed videos. Although large achievements in temporal action proposal generation have been achieved, most previous studies ignore the variability of action frequency in raw videos, leading to unsatisfying performances on high-action-frequency videos. In fact, there exists two main issues which should be well addressed: data imbalance between high and low action-frequency videos, and inferior detection of short actions in high-action-frequency videos. To address the above issues, we propose an effective framework by adapting to the variability of action frequency, namely Action Frequency Adaptive Network (AFAN), which can be flexibly built upon any temporal action proposal generation method. AFAN consists of two modules: Learning From Experts (LFE) and Fine-Grained Processing (FGP). The LFE first trains a series of action proposal generators on different subsets of imbalanced data as experts and then teaches a unified student model via knowledge distillation. To better detect short actions, FGP first finds out high-action-frequency videos and then performs fine-grained detection. Extensive experimental results on four benchmark datasets (ActivityNet-1.3, HACS, THUMOS14 and FineAction) demonstrate the effectiveness and generalizability of the proposed AFAN, especially for high-action-frequency videos.
This paper investigates the use of Neural Network (NN) nonlinear modelling for Power Amplifier (PA) linearization in the Walsh-Hadamard transceiver architecture. This novel architecture has recently been proposed for ultra-high bandwidth systems to reduce the transceiver power consumption by extensive parallelization of the digital baseband hardware. The parallelization is achieved by replacing two-dimensional quadrature modulation with multi-dimensional Walsh-Hadamard modulation. The open research question for this architecture is whether conventional baseband signal processing algorithms can be similarly parallelized while retaining their performance. A key baseband algorithm, digital predistortion using NN models for PA linearization, will be adapted to the parallel Walsh architecture. A straighforward parallelization of the state-of-the-art NN architecture is extended with a cross-domain Knowledge Distillation pre-training method to achieve linearization performance on par with the quadrature implementation. This result paves the way for the entire baseband processing chain to be adapted into ultra-high bandwidth, low-power Walsh transceivers.
The considerable size of Large Language Models (LLMs) presents notable deployment challenges, particularly on resource-constrained hardware. Structured pruning, offers an effective means to compress LLMs, thereby reducing storage costs and enhancing inference speed for more efficient utilization. In this work, we study data-efficient and resource-efficient structure pruning methods to obtain smaller yet still powerful models. Knowledge Distillation is well-suited for pruning, as the intact model can serve as an excellent teacher for pruned students. However, it becomes challenging in the context of LLMs due to memory constraints. To address this, we propose an efficient progressive Numerous-teacher pruning method (NutePrune). NutePrune mitigates excessive memory costs by loading only one intact model and integrating it with various masks and LoRA modules, enabling it to seamlessly switch between teacher and student roles. This approach allows us to leverage numerous teachers with varying capacities to progressively guide the pruned model, enhancing overall performance. Extensive experiments across various tasks demonstrate the effectiveness of NutePrune. In LLaMA-7B zero-shot experiments, NutePrune retains 97.17% of the performance of the original model at 20% sparsity and 95.07% at 25% sparsity. Our code is available at https://github.com/Lucius-lsr/NutePrune.
Transformer based large language models have achieved tremendous success. However, the significant memory and computational costs incurred during the inference process make it challenging to deploy large models on resource-constrained devices. In this paper, we investigate compression and efficient inference methods for large language models from an algorithmic perspective. Regarding taxonomy, similar to smaller models, compression and acceleration algorithms for large language models can still be categorized into quantization, pruning, distillation, compact architecture design, dynamic networks. However, Large language models have two prominent characteristics compared to smaller models: (1) Most of compression algorithms require finetuning or even retraining the model after compression. The most notable aspect of large models is the very high cost associated with model finetuning or training. Therefore, many algorithms for large models, such as quantization and pruning, start to explore tuning-free algorithms. (2) Large models emphasize versatility and generalization rather than performance on a single task. Hence, many algorithms, such as knowledge distillation, focus on how to preserving their versatility and generalization after compression. Since these two characteristics were not very pronounced in early large models, we further distinguish large language models into medium models and ``real'' large models. Additionally, we also provide an introduction to some mature frameworks for efficient inference of large models, which can support basic compression or acceleration algorithms, greatly facilitating model deployment for users.
Emerging Large Language Models (LLMs) like GPT-4 have revolutionized Natural Language Processing (NLP), showing potential in traditional tasks such as Named Entity Recognition (NER). Our study explores a three-phase training strategy that harnesses GPT-4's capabilities to enhance the BERT model's performance on NER. Initially, GPT-4 annotates a subset of the CONLL2003 and additional BBC dataset without fine-tuning. We then train BERT using a mix of original and LLM-annotated data, analyzing the efficacy of LLM annotations against traditional methods. The second phase involves comparative experiments with different training regimens, assessing the synergy between distilled and original data. We observe that sequential strategies, particularly a simple mix of training first with distilled data followed by original data, significantly boost performance. In the third phase, we investigate various data blending techniques, including sigmoid and power decay functions, to optimize the training process further. Our results indicate that a strategic mix of distilled and original data markedly elevates the NER capabilities of BERT. Our approach presents a scalable methodology that reduces manual annotation costs and increases efficiency, making it especially pertinent in resource-limited and closed-network environments. The study concludes that while the 'Simple Mix' strategy yields the best results, understanding its underlying mechanisms requires further research. Future work will also focus on refining prompt designs and enhancing annotation selection processes, aiming to extend our methodology to diverse NLP tasks.
In recent years, federated learning (FL) has emerged as a promising technique for training machine learning models in a decentralized manner while also preserving data privacy. The non-independent and identically distributed (non-i.i.d.) nature of client data, coupled with constraints on client or edge devices, presents significant challenges in FL. Furthermore, learning across a high number of communication rounds can be risky and potentially unsafe for model exploitation. Traditional FL approaches may suffer from these challenges. Therefore, we introduce FedSiKD, which incorporates knowledge distillation (KD) within a similarity-based federated learning framework. As clients join the system, they securely share relevant statistics about their data distribution, promoting intra-cluster homogeneity. This enhances optimization efficiency and accelerates the learning process, effectively transferring knowledge between teacher and student models and addressing device constraints. FedSiKD outperforms state-of-the-art algorithms by achieving higher accuracy, exceeding by 25\% and 18\% for highly skewed data at $\alpha = {0.1,0.5}$ on the HAR and MNIST datasets, respectively. Its faster convergence is illustrated by a 17\% and 20\% increase in accuracy within the first five rounds on the HAR and MNIST datasets, respectively, highlighting its early-stage learning proficiency. Code is publicly available and hosted on GitHub (https://github.com/SimuEnv/FedSiKD)
This study demonstrates the first in-hospital adaptation of a cloud-based AI, similar to ChatGPT, into a secure model for analyzing radiology reports, prioritizing patient data privacy. By employing a unique sentence-level knowledge distillation method through contrastive learning, we achieve over 95% accuracy in detecting anomalies. The model also accurately flags uncertainties in its predictions, enhancing its reliability and interpretability for physicians with certainty indicators. These advancements represent significant progress in developing secure and efficient AI tools for healthcare, suggesting a promising future for in-hospital AI applications with minimal supervision.
Activation function is a pivotal component of deep learning, facilitating the extraction of intricate data patterns. While classical activation functions like ReLU and its variants are extensively utilized, their static nature and simplicity, despite being advantageous, often limit their effectiveness in specialized tasks. The trainable activation functions also struggle sometimes to adapt to the unique characteristics of the data. Addressing these limitations, we introduce a novel trainable activation function, adaptive piecewise approximated activation linear unit (APALU), to enhance the learning performance of deep learning across a broad range of tasks. It presents a unique set of features that enable it to maintain stability and efficiency in the learning process while adapting to complex data representations. Experiments reveal significant improvements over widely used activation functions for different tasks. In image classification, APALU increases MobileNet and GoogleNet accuracy by 0.37% and 0.04%, respectively, on the CIFAR10 dataset. In anomaly detection, it improves the average area under the curve of One-CLASS Deep SVDD by 0.8% on the MNIST dataset, 1.81% and 1.11% improvements with DifferNet, and knowledge distillation, respectively, on the MVTech dataset. Notably, APALU achieves 100% accuracy on a sign language recognition task with a limited dataset. For regression tasks, APALU enhances the performance of deep neural networks and recurrent neural networks on different datasets. These improvements highlight the robustness and adaptability of APALU across diverse deep-learning applications.
Common knowledge distillation methods require the teacher model and the student model to be trained on the same task. However, the usage of embeddings as teachers has also been proposed for different source tasks and target tasks. Prior work that uses embeddings as teachers ignores the fact that the teacher embeddings are likely to contain irrelevant knowledge for the target task. To address this problem, we propose to use an embedding compression module with a trainable teacher transformation to obtain a compact teacher embedding. Results show that adding the embedding compression module improves the classification performance, especially for unsupervised teacher embeddings. Moreover, student models trained with the guidance of embeddings show stronger generalizability.
Medical image segmentation has been significantly advanced by deep learning (DL) techniques, though the data scarcity inherent in medical applications poses a great challenge to DL-based segmentation methods. Self-supervised learning offers a solution by creating auxiliary learning tasks from the available dataset and then leveraging the knowledge acquired from solving auxiliary tasks to help better solve the target segmentation task. Different auxiliary tasks may have different properties and thus can help the target task to different extents. It is desired to leverage their complementary advantages to enhance the overall assistance to the target task. To achieve this, existing methods often adopt a joint training paradigm, which co-solves segmentation and auxiliary tasks by integrating their losses or intermediate gradients. However, direct coupling of losses or intermediate gradients risks undesirable interference because the knowledge acquired from solving each auxiliary task at every training step may not always benefit the target task. To address this issue, we propose a two-stage training approach. In the first stage, the target segmentation task will be independently co-solved with each auxiliary task in both joint training and pre-training modes, with the better model selected via validation performance. In the second stage, the models obtained with respect to each auxiliary task are converted into a single model using an ensemble knowledge distillation method. Our approach allows for making best use of each auxiliary task to create multiple elite segmentation models and then combine them into an even more powerful model. We employed five auxiliary tasks of different proprieties in our approach and applied it to train the U-Net model on an X-ray pneumothorax segmentation dataset. Experimental results demonstrate the superiority of our approach over several existing methods.
In this study, we propose an automated framework for camel farm monitoring, introducing two key contributions: the Unified Auto-Annotation framework and the Fine-Tune Distillation framework. The Unified Auto-Annotation approach combines two models, GroundingDINO (GD), and Segment-Anything-Model (SAM), to automatically annotate raw datasets extracted from surveillance videos. Building upon this foundation, the Fine-Tune Distillation framework conducts fine-tuning of student models using the auto-annotated dataset. This process involves transferring knowledge from a large teacher model to a student model, resembling a variant of Knowledge Distillation. The Fine-Tune Distillation framework aims to be adaptable to specific use cases, enabling the transfer of knowledge from the large models to the small models, making it suitable for domain-specific applications. By leveraging our raw dataset collected from Al-Marmoom Camel Farm in Dubai, UAE, and a pre-trained teacher model, GroundingDINO, the Fine-Tune Distillation framework produces a lightweight deployable model, YOLOv8. This framework demonstrates high performance and computational efficiency, facilitating efficient real-time object detection. Our code is available at \href{https://github.com/Razaimam45/Fine-Tune-Distillation}{https://github.com/Razaimam45/Fine-Tune-Distillation}
Federated Learning (FL) is an emerging machine learning paradigm that enables the collaborative training of a shared global model across distributed clients while keeping the data decentralized. Recent works on designing systems for efficient FL have shown that utilizing serverless computing technologies, particularly Function-as-a-Service (FaaS) for FL, can enhance resource efficiency, reduce training costs, and alleviate the complex infrastructure management burden on data holders. However, existing serverless FL systems implicitly assume a uniform global model architecture across all participating clients during training. This assumption fails to address fundamental challenges in practical FL due to the resource and statistical data heterogeneity among FL clients. To address these challenges and enable heterogeneous client models in serverless FL, we utilize Knowledge Distillation (KD) in this paper. Towards this, we propose novel optimized serverless workflows for two popular conventional federated KD techniques, i.e., FedMD and FedDF. We implement these workflows by introducing several extensions to an open-source serverless FL system called FedLess. Moreover, we comprehensively evaluate the two strategies on multiple datasets across varying levels of client data heterogeneity using heterogeneous client models with respect to accuracy, fine-grained training times, and costs. Results from our experiments demonstrate that serverless FedDF is more robust to extreme non-IID data distributions, is faster, and leads to lower costs than serverless FedMD. In addition, compared to the original implementation, our optimizations for particular steps in FedMD and FedDF lead to an average speedup of 3.5x and 1.76x across all datasets.
Transformer plays a vital role in the realms of natural language processing (NLP) and computer vision (CV), specially for constructing large language models (LLM) and large vision models (LVM). Model compression methods reduce the memory and computational cost of Transformer, which is a necessary step to implement large language/vision models on practical devices. Given the unique architecture of Transformer, featuring alternative attention and feedforward neural network (FFN) modules, specific compression techniques are usually required. The efficiency of these compression methods is also paramount, as retraining large models on the entire training dataset is usually impractical. This survey provides a comprehensive review of recent compression methods, with a specific focus on their application to Transformer-based models. The compression methods are primarily categorized into pruning, quantization, knowledge distillation, and efficient architecture design (Mamba, RetNet, RWKV, etc.). In each category, we discuss compression methods for both language and vision tasks, highlighting common underlying principles. Finally, we delve into the relation between various compression methods, and discuss further directions in this domain.
Knowledge distillation is a simple but powerful way to transfer knowledge between a teacher model to a student model. Existing work suffers from at least one of the following key limitations in terms of direction and scope of transfer which restrict its use: all knowledge is transferred from teacher to student regardless of whether or not that knowledge is useful, the student is the only one learning in this exchange, and typically distillation transfers knowledge only from a single teacher to a single student. We formulate a novel form of knowledge distillation in which many models can act as both students and teachers which we call cooperative distillation. The models cooperate as follows: a model (the student) identifies specific deficiencies in it's performance and searches for another model (the teacher) who encodes learned knowledge into instructional virtual instances via counterfactual instance generation. Because different models may have different strengths and weaknesses, all models can act as either students or teachers (cooperation) when appropriate and only distill knowledge in areas specific to their strengths (focus). Since counterfactuals as a paradigm are not tied to any specific algorithm, we can use this method to distill knowledge between learners of different architectures, algorithms, and even feature spaces. We demonstrate that our approach not only outperforms baselines such as transfer learning, self-supervised learning, and multiple knowledge distillation algorithms on several datasets, but it can also be used in settings where the aforementioned techniques cannot.
Source-free domain adaptation (SFDA) alleviates the domain discrepancy among data obtained from domains without accessing the data for the awareness of data privacy. However, existing conventional SFDA methods face inherent limitations in medical contexts, where medical data are typically collected from multiple institutions using various equipment. To address this problem, we propose a simple yet effective method, named Uncertainty-aware Adaptive Distillation (UAD) for the multi-source-free unsupervised domain adaptation (MSFDA) setting. UAD aims to perform well-calibrated knowledge distillation from (i) model level to deliver coordinated and reliable base model initialisation and (ii) instance level via model adaptation guided by high-quality pseudo-labels, thereby obtaining a high-performance target domain model. To verify its general applicability, we evaluate UAD on two image-based diagnosis benchmarks among two multi-centre datasets, where our method shows a significant performance gain compared with existing works. The code will be available soon.
The advancement of knowledge distillation has played a crucial role in enabling the transfer of knowledge from larger teacher models to smaller and more efficient student models, and is particularly beneficial for online and resource-constrained applications. The effectiveness of the student model heavily relies on the quality of the distilled knowledge received from the teacher. Given the accessibility of unlabelled remote sensing data, semi-supervised learning has become a prevalent strategy for enhancing model performance. However, relying solely on semi-supervised learning with smaller models may be insufficient due to their limited capacity for feature extraction. This limitation restricts their ability to exploit training data. To address this issue, we propose an integrated approach that combines knowledge distillation and semi-supervised learning methods. This hybrid approach leverages the robust capabilities of large models to effectively utilise large unlabelled data whilst subsequently providing the small student model with rich and informative features for enhancement. The proposed semi-supervised learning-based knowledge distillation (SSLKD) approach demonstrates a notable improvement in the performance of the student model, in the application of road segmentation, surpassing the effectiveness of traditional semi-supervised learning methods.
In service-oriented architectures, accurately predicting the Quality of Service (QoS) is crucial for maintaining reliability and enhancing user satisfaction. However, significant challenges remain due to existing methods always overlooking high-order latent collaborative relationships between users and services and failing to dynamically adjust feature learning for every specific user-service invocation, which are critical for learning accurate features. Additionally, reliance on RNNs for capturing QoS evolution hampers models' ability to detect long-term trends due to difficulties in managing long-range dependencies. To address these challenges, we propose the \underline{T}arget-Prompt \underline{O}nline \underline{G}raph \underline{C}ollaborative \underline{L}earning (TOGCL) framework for temporal-aware QoS prediction. TOGCL leverages a dynamic user-service invocation graph to model historical interactions, providing a comprehensive representation of user-service relationships. Building on this graph, it develops a target-prompt graph attention network to extract online deep latent features of users and services at each time slice, simultaneously considering implicit collaborative relationships between target users/services and their neighbors, as well as relevant historical QoS values. Additionally, a multi-layer Transformer encoder is employed to uncover temporal feature evolution patterns of users and services, leading to temporal-aware QoS prediction. Extensive experiments conducted on the WS-DREAM dataset demonstrate that our proposed TOGCL framework significantly outperforms state-of-the-art methods across multiple metrics, achieving improvements of up to 38.80\%. These results underscore the effectiveness of the TOGCL framework for precise temporal QoS prediction.