With the rapid development of deep learning, the increasing complexity and scale of parameters make training a new model increasingly resource-intensive. In this paper, we start from the classic convolutional neural n...
ISBN:
(纸本)9798331314385
With the rapid development of deep learning, the increasing complexity and scale of parameters make training a new model increasingly resource-intensive. In this paper, we start from the classic convolutional neural network (CNN) and explore a paradigm that does not require training to obtain new models. Similar to the birth of CNN inspired by receptive fields in the biological visual system, we draw inspiration from the information subsystem pathways in the biological visual system and propose Model Disassembling and Assembling (MDA). During model disassembling, we introduce the concept of relative contribution and propose a component locating technique to extract task-aware components from trained CNN classifiers. For model assembling, we present the alignment padding strategy and parameter scaling strategy to construct a new model tailored for a specific task, utilizing the disassembled task-aware components. The entire process is akin to playing with LEGO bricks, enabling arbitrary assembly of new models, and providing a novel perspective for model creation and reuse. Extensive experiments showcase that task-aware components disassembled from CNN classifiers or new models assembled using these components closely match or even surpass the performance of the baseline, demonstrating its promising results for model reuse. Furthermore, MDA exhibits diverse potential applications, with comprehensive experiments exploring model decision route analysis, model compression, knowledge distillation, and more. For more information, please visit https://***/.
Early detection and assessment of polyps play a crucial role in the prevention and treatment of colorectal cancer (CRC). Polyp segmentation provides an effective solution to assist clinicians in accurately locating an...
详细信息
Offline reinforcement learning endeavors to leverage offline datasets to craft effective agent policy without online interaction, which imposes proper conservative constraints with the support of behavior policies to ...
ISBN:
(纸本)9798331314385
Offline reinforcement learning endeavors to leverage offline datasets to craft effective agent policy without online interaction, which imposes proper conservative constraints with the support of behavior policies to tackle the out-of-distribution problem. However, existing works often suffer from the constraint conflict issue when offline datasets are collected from multiple behavior policies, i.e., different behavior policies may exhibit inconsistent actions with distinct returns across the state space. To remedy this issue, recent advantage-weighted methods prioritize samples with high advantage values for agent training while inevitably ignoring the diversity of behavior policy. In this paper, we introduce a novel Advantage-Aware Policy Optimization (A2PO) method to explicitly construct advantage-aware policy constraints for offline learning under mixed-quality datasets. Specifically, A2PO employs a conditional variational auto-encoder to disentangle the action distributions of intertwined behavior policies by modeling the advantage values of all training data as conditional variables. Then the agent can follow such disentangled action distribution constraints to optimize the advantage-aware policy towards high advantage values. Extensive experiments conducted on both the single-quality and mixed-quality datasets of the D4RL benchmark demonstrate that A2PO yields results superior to the counterparts. Our code is available at https://***/Plankson/A2PO.
Constraint solving and environment modeling are two challenging problems for symbolic execution. When a program contains non-linear expressions, it is difficult for symbolic execution to explore the program’s whole p...
详细信息
ISBN:
(纸本)9781665455381
Constraint solving and environment modeling are two challenging problems for symbolic execution. When a program contains non-linear expressions, it is difficult for symbolic execution to explore the program’s whole path space due to the high complexity of the constraint solving for the nonlinear constraints. Besides, when the program uses a third-party library and the source code of the library is not available, the symbolic execution of the program often under-approximates the analysis by concrete execution or over-approximates by introducing new symbolic variables, which may fail to explore the whole path space or introduce false alarms, respectively. This paper proposes FUSE, a framework of synergizing symbolic execution and fuzzing by function-level selective symbolization to tackle these problems. First, FUSE collects the path constraints of each function selectively and introduces symbolic function invocation expressions for the complex or third-party functions. Then, FUSE combines SMT solving and fuzzing to solve the path constraints. We have implemented FUSE on the start-of-theart symbolic execution engine KLEE. The experimental results demonstrate that FUSE effectively and efficiently improves the code coverage. Compared with the state-of-the-art, FUSE achieves 6. 6x speedups for achieving the same code coverage.
Deep Neural Network (DNN) Inference, as a key enabler of intelligent applications, is often computation-intensive and latency-sensitive. Combining the advantages of cloud computing (abundant computing resources) and e...
详细信息
Softmax Loss (SL) is widely applied in recommender systems (RS) and has demonstrated effectiveness. This work analyzes SL from a pairwise perspective, revealing two significant limitations: 1) the relationship between...
ISBN:
(纸本)9798331314385
Softmax Loss (SL) is widely applied in recommender systems (RS) and has demonstrated effectiveness. This work analyzes SL from a pairwise perspective, revealing two significant limitations: 1) the relationship between SL and conventional ranking metrics like DCG is not sufficiently tight; 2) SL is highly sensitive to false negative instances. Our analysis indicates that these limitations are primarily due to the use of the exponential function. To address these issues, this work extends SL to a new family of loss functions, termed Pairwise Softmax Loss (PSL), which replaces the exponential function in SL with other appropriate activation functions. While the revision is minimal, we highlight three merits of PSL: 1) it serves as a tighter surrogate for DCG with suitable activation functions; 2) it better balances data contributions; and 3) it acts as a specific BPR loss enhanced by Distributionally Robust Optimization (DRO). We further validate the effectiveness and robustness of PSL through empirical experiments. The code is available at https://***/Tiny-Snow/IR-Benchmark.
We report tunneling spectroscopy of Andreev subgap states in hybrid nanowires with a thin superconducting full shell surrounding a semiconducting core. The combination of the quantized fluxoid of the shell and the And...
详细信息
We report tunneling spectroscopy of Andreev subgap states in hybrid nanowires with a thin superconducting full shell surrounding a semiconducting core. The combination of the quantized fluxoid of the shell and the Andreev reflection at the superconductor-semiconductor interface gives rise to analogs of Caroli–de Gennes–Matricon states found in Abrikosov vortices in type-II superconductors. Unlike in metallic superconductors, Caroli–de Gennes–Matricon analogs in full-shell hybrid nanowires manifest as one-dimensional Van Hove singularities with energy spacings comparable to the superconducting gap and independent of the Fermi energy, making them readily observable. Evolution of these analogs with axial magnetic field, skewed within the Little-Parks lobe structure, is consistent with theory and yields information about the radial distribution and angular momenta of the corresponding subbands.
In this paper, we present a novel surface mesh generation approach based on neural networks and splitting lines that splits B-rep geometry models into isotropic triangular meshes. In the first stage, a recursive metho...
详细信息
In this paper, we present a novel surface mesh generation approach based on neural networks and splitting lines that splits B-rep geometry models into isotropic triangular meshes. In the first stage, a recursive metho...
详细信息
Breast cancer is the most common malignant tumor and the leading cause of cancer-related deaths in women *** means of predicting the prognosis of breast cancer are very helpful in guiding treatment and improving patie...
详细信息
Breast cancer is the most common malignant tumor and the leading cause of cancer-related deaths in women *** means of predicting the prognosis of breast cancer are very helpful in guiding treatment and improving patients'*** extracted by radiomics reflect the genetic and molecular characteristics of a tumor and are related to its biological behavior and the patient's ***,radiomics provides a new approach to noninvasive assessment of breast cancer *** is one of the commonest clinical means of examining breast *** recent years,some results of research into ultrasound radiomics for diagnosing breast cancer,predicting lymph node status,treatment response,recurrence and survival times,and other aspects,have been *** this article,we review the current research status and technical challenges of ultrasound radiomics for predicting breast cancer *** aim to provide a reference for radiomics researchers,promote the development of ultrasound radiomics,and advance its clinical application.
暂无评论