Neural network pruning is useful for discovering efficient, high-performing subnetworks within pre-trained, dense network architectures. More often than not, it involves a three-step process -- pre-training, pruning, and re-training -- that is computationally expensive, as the dense model must be fully pre-trained. While previous work has revealed through experiments the relationship between the amount of pre-training and the performance of the pruned network, a theoretical characterization of such dependency is still missing. Aiming to mathematically analyze the amount of dense network pre-training needed for a pruned network to perform well, we discover a simple theoretical bound in the number of gradient descent pre-training iterations on a two-layer, fully-connected network, beyond which pruning via greedy forward selection [61] yields a subnetwork that achieves good training error. Interestingly, this threshold is shown to be logarithmically dependent upon the size of the dataset, meaning that experiments with larger datasets require more pre-training for subnetworks obtained via pruning to perform well. Lastly, we empirically validate our theoretical results on a multi-layer perceptron trained on MNIST.
相關內容
Networking：IFIP International Conferences on Networking。
Explanation：國(guo)際網絡會議。
Publisher：IFIP。
SIT：
Recently, large pre-trained multilingual speech models have shown potential in scaling Automatic Speech Recognition (ASR) to many low-resource languages. Some of these models employ language adapters in their formulation, which helps to improve monolingual performance and avoids some of the drawbacks of multi-lingual modeling on resource-rich languages. However, this formulation restricts the usability of these models on code-switched speech, where two languages are mixed together in the same utterance. In this work, we propose ways to effectively fine-tune such models on code-switched speech, by assimilating information from both language adapters at each language adaptation point in the network. We also model code-switching as a sequence of latent binary sequences that can be used to guide the flow of information from each language adapter at the frame level. The proposed approaches are evaluated on three code-switched datasets encompassing Arabic, Mandarin, and Hindi languages paired with English, showing consistent improvements in code-switching performance with at least 10\% absolute reduction in CER across all test sets.
Knowledge Graphs (KG) are the backbone of many data-intensive applications since they can represent data coupled with its meaning and context. Aligning KGs across different domains and providers is necessary to afford a fuller and integrated representation. A severe limitation of current KG alignment (KGA) algorithms is that they fail to articulate logical thinking and reasoning with lexical, structural, and semantic data learning. Deep learning models are increasingly popular for KGA inspired by their good performance in other tasks, but they suffer from limitations in explainability, reasoning, and data efficiency. Hybrid neurosymbolic learning models hold the promise of integrating logical and data perspectives to produce high-quality alignments that are explainable and support validation through human-centric approaches. This paper examines the current state of the art in KGA and explores the potential for neurosymbolic integration, highlighting promising research directions for combining these fields.
We propose task-adaptive tokenization as a way to adapt the generation pipeline to the specifics of a downstream task and enhance long-form generation in mental health. Inspired by insights from cognitive science, our task-adaptive tokenizer samples variable segmentations from multiple outcomes, with sampling probabilities optimized based on task-specific data. We introduce a strategy for building a specialized vocabulary and introduce a vocabulary merging protocol that allows for the integration of task-specific tokens into the pre-trained model's tokenization step. Through extensive experiments on psychological question-answering tasks in both Chinese and English, we find that our task-adaptive tokenization approach brings a significant improvement in generation performance while using up to 60% fewer tokens. Preliminary experiments point to promising results when using our tokenization approach with very large language models.
Unlike primates, training artificial neural networks on changing data distributions leads to a rapid decrease in performance on old tasks. This phenomenon is commonly referred to as catastrophic forgetting. In this paper, we investigate the representational changes that underlie this performance decrease and identify three distinct processes that together account for the phenomenon. The largest component is a misalignment between hidden representations and readout layers. Misalignment occurs due to learning on additional tasks and causes internal representations to shift. Representational geometry is partially conserved under this misalignment and only a small part of the information is irrecoverably lost. All types of representational changes scale with the dimensionality of hidden representations. These insights have implications for deep learning applications that need to be continuously updated, but may also aid aligning ANN models to the rather robust biological vision.
Network traffic monitoring based on IP Flows is a standard monitoring approach that can be deployed to various network infrastructures, even the large IPS-based networks connecting millions of people. Since flow records traditionally contain only limited information (addresses, transport ports, and amount of exchanged data), they are also commonly extended for additional features that enable network traffic analysis with high accuracy. Nevertheless, the flow extensions are often too large or hard to compute, which limits their deployment only to smaller-sized networks. This paper proposes a novel extended IP flow called NetTiSA (Network Time Series Analysed), which is based on the analysis of the time series of packet sizes. By thoroughly testing 25 different network classification tasks, we show the broad applicability and high usability of NetTiSA, which often outperforms the best-performing related works. For practical deployment, we also consider the sizes of flows extended for NetTiSA and evaluate the performance impacts of its computation in the flow exporter. The novel feature set proved universal and deployable to high-speed ISP networks with 100\,Gbps lines; thus, it enables accurate and widespread network security protection.
Social media moderation policies are often at the center of public debate, and their implementation and enactment are sometimes surrounded by a veil of mystery. Unsurprisingly, due to limited platform transparency and data access, relatively little research has been devoted to characterizing moderation dynamics, especially in the context of controversial events and the platform activity associated with them. Here, we study the dynamics of account creation and suspension on Twitter during two global political events: Russia's invasion of Ukraine and the 2022 French Presidential election. Leveraging a large-scale dataset of 270M tweets shared by 16M users in multiple languages over several months, we identify peaks of suspicious account creation and suspension, and we characterize behaviours that more frequently lead to account suspension. We show how large numbers of accounts get suspended within days from their creation. Suspended accounts tend to mostly interact with legitimate users, as opposed to other suspicious accounts, often making unwarranted and excessive use of reply and mention features, and predominantly sharing spam and harmful content. While we are only able to speculate about the specific causes leading to a given account suspension, our findings shed light on patterns of platform abuse and subsequent moderation during major events.
Deep neural networks for graphs have emerged as a powerful tool for learning on complex non-euclidean data, which is becoming increasingly common for a variety of different applications. Yet, although their potential has been widely recognised in the machine learning community, graph learning is largely unexplored for downstream tasks such as robotics applications. To fully unlock their potential, hence, we propose a review of graph neural architectures from a robotics perspective. The paper covers the fundamentals of graph-based models, including their architecture, training procedures, and applications. It also discusses recent advancements and challenges that arise in applied settings, related for example to the integration of perception, decision-making, and control. Finally, the paper provides an extensive review of various robotic applications that benefit from learning on graph structures, such as bodies and contacts modelling, robotic manipulation, action recognition, fleet motion planning, and many more. This survey aims to provide readers with a thorough understanding of the capabilities and limitations of graph neural architectures in robotics, and to highlight potential avenues for future research.
Deep neural network based recommendation systems have achieved great success as information filtering techniques in recent years. However, since model training from scratch requires sufficient data, deep learning-based recommendation methods still face the bottlenecks of insufficient data and computational inefficiency. Meta-learning, as an emerging paradigm that learns to improve the learning efficiency and generalization ability of algorithms, has shown its strength in tackling the data sparsity issue. Recently, a growing number of studies on deep meta-learning based recommenddation systems have emerged for improving the performance under recommendation scenarios where available data is limited, e.g. user cold-start and item cold-start. Therefore, this survey provides a timely and comprehensive overview of current deep meta-learning based recommendation methods. Specifically, we propose a taxonomy to discuss existing methods according to recommendation scenarios, meta-learning techniques, and meta-knowledge representations, which could provide the design space for meta-learning based recommendation methods. For each recommendation scenario, we further discuss technical details about how existing methods apply meta-learning to improve the generalization ability of recommendation models. Finally, we also point out several limitations in current research and highlight some promising directions for future research in this area.
The LSTM network was proposed to overcome the difficulty in learning long-term dependence, and has made significant advancements in applications. With its success and drawbacks in mind, this paper raises the question - do RNN and LSTM have long memory? We answer it partially by proving that RNN and LSTM do not have long memory from a statistical perspective. A new definition for long memory networks is further introduced, and it requires the model weights to decay at a polynomial rate. To verify our theory, we convert RNN and LSTM into long memory networks by making a minimal modification, and their superiority is illustrated in modeling long-term dependence of various datasets.
Deep neural networks have achieved remarkable success in computer vision tasks. Existing neural networks mainly operate in the spatial domain with fixed input sizes. For practical applications, images are usually large and have to be downsampled to the predetermined input size of neural networks. Even though the downsampling operations reduce computation and the required communication bandwidth, it removes both redundant and salient information obliviously, which results in accuracy degradation. Inspired by digital signal processing theories, we analyze the spectral bias from the frequency perspective and propose a learning-based frequency selection method to identify the trivial frequency components which can be removed without accuracy loss. The proposed method of learning in the frequency domain leverages identical structures of the well-known neural networks, such as ResNet-50, MobileNetV2, and Mask R-CNN, while accepting the frequency-domain information as the input. Experiment results show that learning in the frequency domain with static channel selection can achieve higher accuracy than the conventional spatial downsampling approach and meanwhile further reduce the input data size. Specifically for ImageNet classification with the same input size, the proposed method achieves 1.41% and 0.66% top-1 accuracy improvements on ResNet-50 and MobileNetV2, respectively. Even with half input size, the proposed method still improves the top-1 accuracy on ResNet-50 by 1%. In addition, we observe a 0.8% average precision improvement on Mask R-CNN for instance segmentation on the COCO dataset.
小貼士
登錄享
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191