Low Rank Decomposition (LRD) is a model compression technique applied to the weight tensors of deep learning models in order to reduce the number of trainable parameters and computational complexity. However, due to high number of new layers added to the architecture after applying LRD, it may not lead to a high training/inference acceleration if the decomposition ranks are not small enough. The issue is that using small ranks increases the risk of significant accuracy drop after decomposition. In this paper, we propose two techniques for accelerating low rank decomposed models without requiring to use small ranks for decomposition. These methods include rank optimization and sequential freezing of decomposed layers. We perform experiments on both convolutional and transformer-based models. Experiments show that these techniques can improve the model throughput up to 60% during training and 37% during inference when combined together while preserving the accuracy close to that of the original models
相關內容
Federated Learning (FL) is a promising research paradigm that enables the collaborative training of machine learning models among various parties without the need for sensitive information exchange. Nonetheless, retaining data in individual clients introduces fundamental challenges to achieving performance on par with centrally trained models. Our study provides an extensive review of federated learning applied to visual recognition. It underscores the critical role of thoughtful architectural design choices in achieving optimal performance, a factor often neglected in the FL literature. Many existing FL solutions are tested on shallow or simple networks, which may not accurately reflect real-world applications. This practice restricts the transferability of research findings to large-scale visual recognition models. Through an in-depth analysis of diverse cutting-edge architectures such as convolutional neural networks, transformers, and MLP-mixers, we experimentally demonstrate that architectural choices can substantially enhance FL systems' performance, particularly when handling heterogeneous data. We study 19 visual recognition models from five different architectural families on four challenging FL datasets. We also re-investigate the inferior performance of convolution-based architectures in the FL setting and analyze the influence of normalization layers on the FL performance. Our findings emphasize the importance of architectural design for computer vision tasks in practical scenarios, effectively narrowing the performance gap between federated and centralized learning. Our source code is available at //github.com/sarapieri/fed_het.git.
Federated learning (FL) is a promising paradigm to enable collaborative model training with decentralized data. However, the training process of Large Language Models (LLMs) generally incurs the update of significant parameters, which limits the applicability of FL techniques to tackle the LLMs in real scenarios. Prompt tuning can significantly reduce the number of parameters to update, but it either incurs performance degradation or low training efficiency. The straightforward utilization of prompt tuning in the FL often raises non-trivial communication costs and dramatically degrades performance. In addition, the decentralized data is generally non-Independent and Identically Distributed (non-IID), which brings client drift problems and thus poor performance. This paper proposes a Parameter-efficient prompt Tuning approach with Adaptive Optimization, i.e., FedPepTAO, to enable efficient and effective FL of LLMs. First, an efficient partial prompt tuning approach is proposed to improve performance and efficiency simultaneously. Second, a novel adaptive optimization method is developed to address the client drift problems on both the device and server sides to enhance performance further. Extensive experiments based on 10 datasets demonstrate the superb performance (up to 60.8\% in terms of accuracy) and efficiency (up to 97.59\% in terms of training time) of FedPepTAO compared with 9 baseline approaches. Our code is available at //github.com/llm-eff/FedPepTAO.
The quality and quantity of data used for training greatly influence the performance and effectiveness of deep learning models. In the context of error correction, it is essential to generate high-quality samples that are neither excessively noisy nor entirely correct but close to the decoding region's decision boundary. To accomplish this objective, this paper utilizes a restricted version of a recent result on Importance Sampling (IS) distribution for fast performance evaluation of linear codes. The IS distribution is used over the segmented observation space and integrated with active learning. This combination allows for the iterative generation of samples from the shells whose acquisition functions, defined as the error probabilities conditioned on each shell, fall within a specific range. By intelligently sampling based on the proposed IS distribution, significant improvements are demonstrated in the performance of BCH(63,36) and BCH(63,45) codes with cycle-reduced parity-check matrices. The proposed IS-based-active Weight Belief Propagation (WBP) decoder shows improvements of up to 0.4dB in the waterfall region and up to 1.9dB in the error-floor region of the BER curve, over the conventional WBP. This approach can be easily adapted to generate efficient samples to train any other deep learning-based decoder.
Among the wide variety of evolutionary computing models, Finite State Machines (FSMs) have several attractions for fundamental research. They are easy to understand in concept and can be visualised clearly in simple cases. They have a ready fitness criterion through their relationship with Regular Languages. They have also been shown to be tractably evolvable, even up to exhibiting evidence of open-ended evolution in specific scenarios. In addition to theoretical attraction, they also have industrial applications, as a paradigm of both automated and user-initiated control. Improving the understanding of the factors affecting FSM evolution has relevance to both computer science and practical optimisation of control. We investigate an evolutionary scenario of FSMs adapting to recognise one of a family of Regular Languages by categorising positive and negative samples, while also being under a counteracting selection pressure that favours fewer states. The results appear to indicate that longer strings provided as samples reduce the speed of fitness gain, when fitness is measured against a fixed number of sample strings. We draw the inference that additional information from longer strings is not sufficient to compensate for sparser coverage of the combinatorial space of positive and negative sample strings.
Physics Informed Neural Networks (PINNs) represent the intersection between physics-based modeling and deep learning, but successfully training PINNs in 3D for highly nonlinear PDEs on complex domains remains a challenging task. In this paper, PINNs are used to solve the 3D incompressible Navier-Stokes (NS) equations at high Reynolds numbers for complex geometries, using very sparsely distributed solution data in the domain. The effect of the amount of data provided and the PDE-based regularizers are investigated. Additionally, hybrid data-PINNs are used to create surrogate models to solve a realistic flow-thermal electronics design problem in near real-time, and it is found that the hybrid data-PINNs consistently outperform standard data-driven neural networks when tested on unseen query points. The findings of the paper show how PINNs can be effective when used in conjunction with sparse data for solving 3D nonlinear PDEs or for surrogate modeling of design spaces governed by them.
Within the realm of deep learning, the interpretability of Convolutional Neural Networks (CNNs), particularly in the context of image classification tasks, remains a formidable challenge. To this end we present a neurosymbolic framework, NeSyFOLD-G that generates a symbolic rule-set using the last layer kernels of the CNN to make its underlying knowledge interpretable. What makes NeSyFOLD-G different from other similar frameworks is that we first find groups of similar kernels in the CNN (kernel-grouping) using the cosine-similarity between the feature maps generated by various kernels. Once such kernel groups are found, we binarize each kernel group's output in the CNN and use it to generate a binarization table which serves as input data to FOLD-SE-M which is a Rule Based Machine Learning (RBML) algorithm. FOLD-SE-M then generates a rule-set that can be used to make predictions. We present a novel kernel grouping algorithm and show that grouping similar kernels leads to a significant reduction in the size of the rule-set generated by FOLD-SE-M, consequently, improving the interpretability. This rule-set symbolically encapsulates the connectionist knowledge of the trained CNN. The rule-set can be viewed as a normal logic program wherein each predicate's truth value depends on a kernel group in the CNN. Each predicate in the rule-set is mapped to a concept using a few semantic segmentation masks of the images used for training, to make it human-understandable. The last layers of the CNN can then be replaced by this rule-set to obtain the NeSy-G model which can then be used for the image classification task. The goal directed ASP system s(CASP) can be used to obtain the justification of any prediction made using the NeSy-G model. We also propose a novel algorithm for labeling each predicate in the rule-set with the semantic concept(s) that its corresponding kernel group represents.
Multivariate Analysis on Performance Gaps of Artificial Intelligence Models in Screening Mammography
Although deep learning models for abnormality classification can perform well in screening mammography, the demographic, imaging, and clinical characteristics associated with increased risk of model failure remain unclear. This retrospective study uses the Emory BrEast Imaging Dataset(EMBED) containing mammograms from 115931 patients imaged at Emory Healthcare between 2013-2020, with BI-RADS assessment, region of interest coordinates for abnormalities, imaging features, pathologic outcomes, and patient demographics. Multiple deep learning models were trained to distinguish between abnormal tissue patches and randomly selected normal tissue patches from screening mammograms. We assessed model performance by subgroups defined by age, race, pathologic outcome, tissue density, and imaging characteristics and investigated their associations with false negatives (FN) and false positives (FP). We also performed multivariate logistic regression to control for confounding between subgroups. The top-performing model, ResNet152V2, achieved accuracy of 92.6%(95%CI=92.0-93.2%), and AUC 0.975(95%CI=0.972-0.978). Before controlling for confounding, nearly all subgroups showed statistically significant differences in model performance. However, after controlling for confounding, we found lower FN risk associates with Other race(RR=0.828;p=.050), biopsy-proven benign lesions(RR=0.927;p=.011), and mass(RR=0.921;p=.010) or asymmetry(RR=0.854;p=.040); higher FN risk associates with architectural distortion (RR=1.037;p<.001). Higher FP risk associates to BI-RADS density C(RR=1.891;p<.001) and D(RR=2.486;p<.001). Our results demonstrate subgroup analysis is important in mammogram classifier performance evaluation, and controlling for confounding between subgroups elucidates the true associations between variables and model failure. These results can help guide developing future breast cancer detection models.
Compressing neural networks is a key step when deploying models for real-time or embedded applications. Factorizing the model's matrices using low-rank approximations is a promising method for achieving compression. While it is possible to set the rank before training, this approach is neither flexible nor optimal. In this work, we propose a post-training rank-selection method called Rank-Tuning that selects a different rank for each matrix. Used in combination with training adaptations, our method achieves high compression rates with no or little performance degradation. Our numerical experiments on signal processing tasks show that we can compress recurrent neural networks up to 14x with at most 1.4% relative performance reduction.
Pre-trained Language Models (PLMs) which are trained on large text corpus via self-supervised learning method, have yielded promising performance on various tasks in Natural Language Processing (NLP). However, though PLMs with huge parameters can effectively possess rich knowledge learned from massive training text and benefit downstream tasks at the fine-tuning stage, they still have some limitations such as poor reasoning ability due to the lack of external knowledge. Research has been dedicated to incorporating knowledge into PLMs to tackle these issues. In this paper, we present a comprehensive review of Knowledge-Enhanced Pre-trained Language Models (KE-PLMs) to provide a clear insight into this thriving field. We introduce appropriate taxonomies respectively for Natural Language Understanding (NLU) and Natural Language Generation (NLG) to highlight these two main tasks of NLP. For NLU, we divide the types of knowledge into four categories: linguistic knowledge, text knowledge, knowledge graph (KG), and rule knowledge. The KE-PLMs for NLG are categorized into KG-based and retrieval-based methods. Finally, we point out some promising future directions of KE-PLMs.
Federated learning (FL) is an emerging, privacy-preserving machine learning paradigm, drawing tremendous attention in both academia and industry. A unique characteristic of FL is heterogeneity, which resides in the various hardware specifications and dynamic states across the participating devices. Theoretically, heterogeneity can exert a huge influence on the FL training process, e.g., causing a device unavailable for training or unable to upload its model updates. Unfortunately, these impacts have never been systematically studied and quantified in existing FL literature. In this paper, we carry out the first empirical study to characterize the impacts of heterogeneity in FL. We collect large-scale data from 136k smartphones that can faithfully reflect heterogeneity in real-world settings. We also build a heterogeneity-aware FL platform that complies with the standard FL protocol but with heterogeneity in consideration. Based on the data and the platform, we conduct extensive experiments to compare the performance of state-of-the-art FL algorithms under heterogeneity-aware and heterogeneity-unaware settings. Results show that heterogeneity causes non-trivial performance degradation in FL, including up to 9.2% accuracy drop, 2.32x lengthened training time, and undermined fairness. Furthermore, we analyze potential impact factors and find that device failure and participant bias are two potential factors for performance degradation. Our study provides insightful implications for FL practitioners. On the one hand, our findings suggest that FL algorithm designers consider necessary heterogeneity during the evaluation. On the other hand, our findings urge system providers to design specific mechanisms to mitigate the impacts of heterogeneity.
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