Overparametrized transformer networks are the state-of-the-art architecture for Large Language Models (LLMs). However, such models contain billions of parameters making large compute a necessity, while raising environmental concerns. To address these issues, we propose FinerCut, a new form of fine-grained layer pruning, which in contrast to prior work at the transformer block level, considers all self-attention and feed-forward network (FFN) layers within blocks as individual pruning candidates. FinerCut prunes layers whose removal causes minimal alternation to the model's output -- contributing to a new, lean, interpretable, and task-agnostic pruning method. Tested across 9 benchmarks, our approach retains 90% performance of Llama3-8B with 25% layers removed, and 95% performance of Llama3-70B with 30% layers removed, all without fine-tuning or post-pruning reconstruction. Strikingly, we observe intriguing results with FinerCut: 42% (34 out of 80) of the self-attention layers in Llama3-70B can be removed while preserving 99% of its performance -- without additional fine-tuning after removal. Moreover, FinerCut provides a tool to inspect the types and locations of pruned layers, allowing to observe interesting pruning behaviors. For instance, we observe a preference for pruning self-attention layers, often at deeper consecutive decoder layers. We hope our insights inspire future efficient LLM architecture designs.
相關內容
The depth/thermal information is beneficial for detecting salient object with conventional RGB images. However, in dual-modal salient object detection (SOD) model, the robustness against noisy inputs and modality missing is crucial but rarely studied. To tackle this problem, we introduce \textbf{Co}nditional Dropout and \textbf{LA}nguage-driven(\textbf{CoLA}) framework comprising two core components. 1) Language-driven Quality Assessment (LQA): Leveraging a pretrained vision-language model with a prompt learner, the LQA recalibrates image contributions without requiring additional quality annotations. This approach effectively mitigates the impact of noisy inputs. 2) Conditional Dropout (CD): A learning method to strengthen the model's adaptability in scenarios with missing modalities, while preserving its performance under complete modalities. The CD serves as a plug-in training scheme that treats modality-missing as conditions, strengthening the overall robustness of various dual-modal SOD models. Extensive experiments demonstrate that the proposed method outperforms state-of-the-art dual-modal SOD models, under both modality-complete and modality-missing conditions. We will release source code upon acceptance.
Software Defined Networking (SDN) has brought significant advancements in network management and programmability. However, this evolution has also heightened vulnerability to Advanced Persistent Threats (APTs), sophisticated and stealthy cyberattacks that traditional detection methods often fail to counter, especially in the face of zero-day exploits. A prevalent issue is the inadequacy of existing strategies to detect novel threats while addressing data privacy concerns in collaborative learning scenarios. This paper presents P3GNN (privacy-preserving provenance graph-based graph neural network model), a novel model that synergizes Federated Learning (FL) with Graph Convolutional Networks (GCN) for effective APT detection in SDN environments. P3GNN utilizes unsupervised learning to analyze operational patterns within provenance graphs, identifying deviations indicative of security breaches. Its core feature is the integration of FL with homomorphic encryption, which fortifies data confidentiality and gradient integrity during collaborative learning. This approach addresses the critical challenge of data privacy in shared learning contexts. Key innovations of P3GNN include its ability to detect anomalies at the node level within provenance graphs, offering a detailed view of attack trajectories and enhancing security analysis. Furthermore, the models unsupervised learning capability enables it to identify zero-day attacks by learning standard operational patterns. Empirical evaluation using the DARPA TCE3 dataset demonstrates P3GNNs exceptional performance, achieving an accuracy of 0.93 and a low false positive rate of 0.06.
We introduce MMIS, a novel dataset designed to advance MultiModal Interior Scene generation and recognition. MMIS consists of nearly 160,000 images. Each image within the dataset is accompanied by its corresponding textual description and an audio recording of that description, providing rich and diverse sources of information for scene generation and recognition. MMIS encompasses a wide range of interior spaces, capturing various styles, layouts, and furnishings. To construct this dataset, we employed careful processes involving the collection of images, the generation of textual descriptions, and corresponding speech annotations. The presented dataset contributes to research in multi-modal representation learning tasks such as image generation, retrieval, captioning, and classification.
Multicasting is a vital information dissemination technique in Software-Defined Networking (SDN). With SDN, a multicast service can incorporate network functions implemented at different nodes, which is referred to as software-defined multicast. Emerging ubiquitous wireless networks for 5G and Beyond (B5G) inherently support multicast. However, the broadcast nature of wireless channels, especially in dense deployments, leads to neighborhood interference as a primary system degradation factor, which introduces a new challenge for software-defined multicast in wireless mesh networks. To tackle this, this paper introduces an innovative approach, based on the idea of minimizing both the total length cost of the multicast tree and the interference at the same time. Accordingly, a novel bicriteria optimization problem is formulated--\emph{Minimum Interference Steiner Tree (MIST)}, which is the edge-weighted variant of the vertex-weighted secluded Steiner tree problem \cite{chechik2013secluded}. To solve the bicriteria problem, instead of resorting to heuristics, this paper employs an innovative approach that is an approximate algorithm for MIST but with guaranteed performance. Specifically, the approach exploits the monotone submodularity property of the interference metric and identifies Pareto optimal solutions for MIST, then converts the problem into the submodular minimization under Steiner tree constraints, and designs a two-stage relaxation algorithm. Simulation results demonstrate and validate the performance of the proposed algorithm.
DEVS/SOA: A Cross-Platform Framework for Net-centric Modeling and Simulation in DEVS Unified Process
Discrete EVent Specification (DEVS) environments are known to be implemented over middleware systems such as HLA, RMI, CORBA and others. DEVS exhibits concepts of systems theory and modeling and supports capturing the system behavior from the physical and behavioral perspectives. Further, they are implemented using Object-oriented languages like Java and C++. This research work uses the Java platform to implement DEVS over a Service Oriented Architecture (SOA) framework. Called the DEVS/SOA, the framework supports a development and testing environment known as DEVS Unified Process that is built on a model-continuity-based life cycle methodology. DEVS Unified Process allows DEVS-based Modeling and Simulation (M&S) over net-centric platforms using DEVS/SOA. This framework also provides the crucial feature of run-time composability of coupled systems using SOA. We describe the architecture and designs of the both the server and the client. The client application communicates with multiple servers hosting DEVS simulation services. These Simulation services are developed using the proposed symmetrical services architecture wherein the server can act as both a service provider and a service consumer contrary to the unidirectional client-server paradigm. We also discuss how this Services based architecture provides solutions for cross-platform distributed M&S. We demonstrate DEVS/SOA framework with a scenario of Joint Close Air Support specified in Business Process Modeling Notation (BPMN). We also provide a real-world application of Network health monitoring using DEVS/SOA layered architectural framework.
With the urgent demand for generalized deep models, many pre-trained big models are proposed, such as BERT, ViT, GPT, etc. Inspired by the success of these models in single domains (like computer vision and natural language processing), the multi-modal pre-trained big models have also drawn more and more attention in recent years. In this work, we give a comprehensive survey of these models and hope this paper could provide new insights and helps fresh researchers to track the most cutting-edge works. Specifically, we firstly introduce the background of multi-modal pre-training by reviewing the conventional deep learning, pre-training works in natural language process, computer vision, and speech. Then, we introduce the task definition, key challenges, and advantages of multi-modal pre-training models (MM-PTMs), and discuss the MM-PTMs with a focus on data, objectives, network architectures, and knowledge enhanced pre-training. After that, we introduce the downstream tasks used for the validation of large-scale MM-PTMs, including generative, classification, and regression tasks. We also give visualization and analysis of the model parameters and results on representative downstream tasks. Finally, we point out possible research directions for this topic that may benefit future works. In addition, we maintain a continuously updated paper list for large-scale pre-trained multi-modal big models: //github.com/wangxiao5791509/MultiModal_BigModels_Survey
Knowledge enhanced pre-trained language models (K-PLMs) are shown to be effective for many public tasks in the literature but few of them have been successfully applied in practice. To address this problem, we propose K-AID, a systematic approach that includes a low-cost knowledge acquisition process for acquiring domain knowledge, an effective knowledge infusion module for improving model performance, and a knowledge distillation component for reducing the model size and deploying K-PLMs on resource-restricted devices (e.g., CPU) for real-world application. Importantly, instead of capturing entity knowledge like the majority of existing K-PLMs, our approach captures relational knowledge, which contributes to better-improving sentence-level text classification and text matching tasks that play a key role in question answering (QA). We conducted a set of experiments on five text classification tasks and three text matching tasks from three domains, namely E-commerce, Government, and Film&TV, and performed online A/B tests in E-commerce. Experimental results show that our approach is able to achieve substantial improvement on sentence-level question answering tasks and bring beneficial business value in industrial settings.
Recently, the emergence of pre-trained models (PTMs) has brought natural language processing (NLP) to a new era. In this survey, we provide a comprehensive review of PTMs for NLP. We first briefly introduce language representation learning and its research progress. Then we systematically categorize existing PTMs based on a taxonomy with four perspectives. Next, we describe how to adapt the knowledge of PTMs to the downstream tasks. Finally, we outline some potential directions of PTMs for future research. This survey is purposed to be a hands-on guide for understanding, using, and developing PTMs for various NLP tasks.
Most existing event extraction (EE) methods merely extract event arguments within the sentence scope. However, such sentence-level EE methods struggle to handle soaring amounts of documents from emerging applications, such as finance, legislation, health, etc., where event arguments always scatter across different sentences, and even multiple such event mentions frequently co-exist in the same document. To address these challenges, we propose a novel end-to-end model, Doc2EDAG, which can generate an entity-based directed acyclic graph to fulfill the document-level EE (DEE) effectively. Moreover, we reformalize a DEE task with the no-trigger-words design to ease the document-level event labeling. To demonstrate the effectiveness of Doc2EDAG, we build a large-scale real-world dataset consisting of Chinese financial announcements with the challenges mentioned above. Extensive experiments with comprehensive analyses illustrate the superiority of Doc2EDAG over state-of-the-art methods. Data and codes can be found at //github.com/dolphin-zs/Doc2EDAG.
We introduce a new language representation model called BERT, which stands for Bidirectional Encoder Representations from Transformers. Unlike recent language representation models, BERT is designed to pre-train deep bidirectional representations from unlabeled text by jointly conditioning on both left and right context in all layers. As a result, the pre-trained BERT model can be fine-tuned with just one additional output layer to create state-of-the-art models for a wide range of tasks, such as question answering and language inference, without substantial task-specific architecture modifications. BERT is conceptually simple and empirically powerful. It obtains new state-of-the-art results on eleven natural language processing tasks, including pushing the GLUE score to 80.5% (7.7% point absolute improvement), MultiNLI accuracy to 86.7% (4.6% absolute improvement), SQuAD v1.1 question answering Test F1 to 93.2 (1.5 point absolute improvement) and SQuAD v2.0 Test F1 to 83.1 (5.1 point absolute improvement).
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191