All tables can be represented as grids. Based on this observation, we propose GridFormer, a novel approach for interpreting unconstrained table structures by predicting the vertex and edge of a grid. First, we propose a flexible table representation in the form of an MXN grid. In this representation, the vertexes and edges of the grid store the localization and adjacency information of the table. Then, we introduce a DETR-style table structure recognizer to efficiently predict this multi-objective information of the grid in a single shot. Specifically, given a set of learned row and column queries, the recognizer directly outputs the vertexes and edges information of the corresponding rows and columns. Extensive experiments on five challenging benchmarks which include wired, wireless, multi-merge-cell, oriented, and distorted tables demonstrate the competitive performance of our model over other methods.
Deep Neural Networks (DNNs) have been a large driver and enabler for AI breakthroughs in recent years. These models have been getting larger in their attempt to become more accurate and tackle new upcoming use-cases, including AR/VR and intelligent assistants. However, the training process of such large models is a costly and time-consuming process, which typically yields a single model to fit all targets. To mitigate this, various techniques have been proposed in the literature, including pruning, sparsification or quantization of the model weights and updates. While able to achieve high compression rates, they often incur computational overheads or accuracy penalties. Alternatively, factorization methods have been leveraged to incorporate low-rank compression in the training process. Similarly, such techniques (e.g.,~SVD) frequently rely on the computationally expensive decomposition of layers and are potentially sub-optimal for non-linear models, such as DNNs. In this work, we take a further step in designing efficient low-rank models and propose Maestro, a framework for trainable low-rank layers. Instead of regularly applying a priori decompositions such as SVD, the low-rank structure is built into the training process through a generalized variant of Ordered Dropout. This method imposes an importance ordering via sampling on the decomposed DNN structure. Our theoretical analysis demonstrates that our method recovers the SVD decomposition of linear mapping on uniformly distributed data and PCA for linear autoencoders. We further apply our technique on DNNs and empirically illustrate that Maestro enables the extraction of lower footprint models that preserve model performance while allowing for graceful accuracy-latency tradeoff for the deployment to devices of different capabilities.
An optoelectronic oscillation method with reconfigurable multiple formats for simultaneous generation of coherent dual-band signals is proposed and experimentally demonstrated. By introducing a compatible filtering mechanism based on stimulated Brillouin scattering (SBS) effect into a typical Phase-shifted grating Bragg fiber (PS-FBG) notch filtering cavity, dual mode-selection mechanisms which have independent frequency and time tuning mechanism can be constructed. By regulating three controllers, the proposed scheme can work in different states, named mode 1, mode 2 and mode 3. At mode 1 state, a dual single-frequency hopping signals is achieved with 50 ns hopping speed and flexible central frequency and pulse duration ratio. The mode 2 state is realized by applying the Fourier domain mode-locked (FDML) technology into the proposed optoelectrical oscillator, in which dual coherent pulsed single-frequency signal and broadband signal is generated simultaneously. The adjustability of the time duration of the single-frequency signal and the bandwidth of the broadband signal are shown and discussed. The mode 3 state is a dual broadband signal generator which is realized by injecting a triangular wave into the signal laser. The detection performance of the generated broadband signals has also been evaluated by the pulse compression and the phase noise figure. The proposed method may provide a multifunctional radar system signal generator based on the simply external controllers, which can realize low-phase-noise or multifunctional detection with high resolution imaging ability, especially in a complex interference environment.
Recent research has shown that training low-rank neural networks can effectively reduce the total number of trainable parameters without sacrificing predictive accuracy, resulting in end-to-end speedups. However, low-rank model training necessitates adjusting several additional factorization hyperparameters, such as the rank of the factorization at each layer. In this paper, we tackle this challenge by introducing Cuttlefish, an automated low-rank training approach that eliminates the need for tuning factorization hyperparameters. Cuttlefish leverages the observation that after a few epochs of full-rank training, the stable rank (i.e., an approximation of the true rank) of each layer stabilizes at a constant value. Cuttlefish switches from full-rank to low-rank training once the stable ranks of all layers have converged, setting the dimension of each factorization to its corresponding stable rank. Our results show that Cuttlefish generates models up to 5.6 times smaller than full-rank models, and attains up to a 1.2 times faster end-to-end training process while preserving comparable accuracy. Moreover, Cuttlefish outperforms state-of-the-art low-rank model training methods and other prominent baselines. The source code for our implementation can be found at: https://github.com/hwang595/Cuttlefish.
Whole Slide Image (WSI) classification remains a challenge due to their extremely high resolution and the absence of fine-grained labels. Presently, WSIs are usually classified as a Multiple Instance Learning (MIL) problem when only slide-level labels are available. MIL methods involve a patch embedding process and a bag-level classification process, but they are prohibitively expensive to be trained end-to-end. Therefore, existing methods usually train them separately, or directly skip the training of the embedder. Such schemes hinder the patch embedder's access to slide-level labels, resulting in inconsistencies within the entire MIL pipeline. To overcome this issue, we propose a novel framework called Iteratively Coupled MIL (ICMIL), which bridges the loss back-propagation process from the bag-level classifier to the patch embedder. In ICMIL, we use category information in the bag-level classifier to guide the patch-level fine-tuning of the patch feature extractor. The refined embedder then generates better instance representations for achieving a more accurate bag-level classifier. By coupling the patch embedder and bag classifier at a low cost, our proposed framework enables information exchange between the two processes, benefiting the entire MIL classification model. We tested our framework on two datasets using three different backbones, and our experimental results demonstrate consistent performance improvements over state-of-the-art MIL methods. Code will be made available upon acceptance.
Federated learning is a promising paradigm that allows multiple clients to collaboratively train a model without sharing the local data. However, the presence of heterogeneous devices in federated learning, such as mobile phones and IoT devices with varying memory capabilities, would limit the scale and hence the performance of the model could be trained. The mainstream approaches to address memory limitations focus on width-slimming techniques, where different clients train subnetworks with reduced widths locally and then the server aggregates the subnetworks. The global model produced from these methods suffers from performance degradation due to the negative impact of the actions taken to handle the varying subnetwork widths in the aggregation phase. In this paper, we introduce a memory-adaptive depth-wise learning solution in FL called FeDepth, which adaptively decomposes the full model into blocks according to the memory budgets of each client and trains blocks sequentially to obtain a full inference model. Our method outperforms state-of-the-art approaches, achieving 5% and more than 10% improvements in top-1 accuracy on CIFAR-10 and CIFAR-100, respectively. We also demonstrate the effectiveness of depth-wise fine-tuning on ViT. Our findings highlight the importance of memory-aware techniques for federated learning with heterogeneous devices and the success of depth-wise training strategy in improving the global model's performance.
The canonical formulation of federated learning treats it as a distributed optimization problem where the model parameters are optimized against a global loss function that decomposes across client loss functions. A recent alternative formulation instead treats federated learning as a distributed inference problem, where the goal is to infer a global posterior from partitioned client data (Al-Shedivat et al., 2021). This paper extends the inference view and describes a variational inference formulation of federated learning where the goal is to find a global variational posterior that well-approximates the true posterior. This naturally motivates an expectation propagation approach to federated learning (FedEP), where approximations to the global posterior are iteratively refined through probabilistic message-passing between the central server and the clients. We conduct an extensive empirical study across various algorithmic considerations and describe practical strategies for scaling up expectation propagation to the modern federated setting. We apply FedEP on standard federated learning benchmarks and find that it outperforms strong baselines in terms of both convergence speed and accuracy.
Large-scale Transformer models are known for their exceptional performance in a range of tasks, but training them can be difficult due to the requirement for communication-intensive model parallelism. One way to improve training speed is to compress the message size in communication. Previous approaches have primarily focused on compressing gradients in a data parallelism setting, but compression in a model-parallel setting is an understudied area. We have discovered that model parallelism has fundamentally different characteristics than data parallelism. In this work, we present the first empirical study on the effectiveness of compression methods for model parallelism. We implement and evaluate three common classes of compression algorithms - pruning-based, learning-based, and quantization-based - using a popular Transformer training framework. We evaluate these methods across more than 160 settings and 8 popular datasets, taking into account different hyperparameters, hardware, and both fine-tuning and pre-training stages. We also provide analysis when the model is scaled up. Finally, we provide insights for future development of model parallelism compression algorithms.
Enabling private inference is crucial for many cloud inference services that are based on Transformer models. However, existing private inference solutions for Transformers can increase the inference latency by more than 60x or significantly compromise the quality of inference results. In this paper, we design the framework MPCFORMER using secure multi-party computation (MPC) and Knowledge Distillation (KD). It can be used in tandem with many specifically designed MPC-friendly approximations and trained Transformer models. MPCFORMER significantly speeds up Transformer model inference in MPC settings while achieving similar ML performance to the input model. We evaluate MPCFORMER with various settings in MPC. On the IMDb dataset, we achieve similar performance to BERTBASE, while being 5.3x faster. On the GLUE benchmark, we achieve 97% performance of BERTBASE with a 2.2x speedup. We show that MPCFORMER remains effective with different trained Transformer weights such as ROBERTABASE and larger models including BERTLarge. In particular, we achieve similar performance to BERTLARGE, while being 5.93x faster on the IMDb dataset.
High resolution (HR) 3D medical image segmentation plays an important role in clinical diagnoses. However, HR images are difficult to be directly processed by mainstream graphical cards due to limited video memory. Therefore, most existing 3D medical image segmentation methods use patch-based models, which ignores global context information that is useful in accurate segmentation and has low inference efficiency. To address these problems, we propose a super-resolution (SR) guided patch-free 3D medical image segmentation framework that can realize HR segmentation with global information of low-resolution (LR) input. The framework contains two tasks: semantic segmentation (main task) and super resolution (auxiliary task). To balance the information loss with the LR input, we introduce a Self-Supervised Guidance Module (SGM), which employs a selective search method to crop a HR patch from the original image as restoration guidance. Multi-scale convolutional layers are used to mitigate the scale-inconsistency between the HR guidance features and the LR features. Moreover, we propose a Task-Fusion Module (TFM) to exploit the inter connections between segmentation and SR task. This module can also be used for Test Phase Fine-tuning (TPF), leading to a better model generalization ability. When predicting, only the main segmentation task is needed, while other modules can be removed to accelerate the inference. The experiments results on two different datasets show that our framework outperforms current patch-based and patch-free models. Our model also has a four times higher inference speed compared to traditional patch-based methods. Our codes are available at: https://github.com/Dootmaan/PFSeg-Full.