Alpa: Automating Inter- and Intra-Operator Parallelism for Distributed Deep Learning

Alpa automates model-parallel training of large deep learning (DL) models by generating execution plans that unify data, operator, and pipeline parallelism. Existing model-parallel training systems either require users to manually create a parallelization plan or automatically generate one from a limited space of model parallelism configurations, which does not suffice to scale out complex DL models on distributed compute devices. Alpa distributes the training of large DL models by viewing parallelisms as two hierarchical levels: inter-operator and intra-operator parallelisms. Based on it, Alpa constructs a new hierarchical space for massive model-parallel execution plans. Alpa designs a number of compilation passes to automatically derive the optimal parallel execution plan in each independent parallelism level and implements an efficient runtime to orchestrate the two-level parallel execution on distributed compute devices. Our evaluation shows Alpa generates parallelization plans that match or outperform hand-tuned model-parallel training systems even on models they are designed for. Unlike specialized systems, Alpa also generalizes to models with heterogeneous architectures and models without manually-designed plans.

The Elements of Temporal Sentence Grounding in Videos: A Survey and Future Directions

Temporal sentence grounding in videos (TSGV), a.k.a., natural language video localization (NLVL) or video moment retrieval (VMR), aims to retrieve a temporal moment that semantically corresponds to a language query from an untrimmed video. Connecting computer vision and natural language, TSGV has drawn significant attention from researchers in both communities. This survey attempts to provide a summary of fundamental concepts in TSGV and current research status, as well as future research directions. As the background, we present a common structure of functional components in TSGV, in a tutorial style: from feature extraction from raw video and language query, to answer prediction of the target moment. Then we review the techniques for multimodal understanding and interaction, which is the key focus of TSGV for effective alignment between the two modalities. We construct a taxonomy of TSGV techniques and elaborate methods in different categories with their strengths and weaknesses. Lastly, we discuss issues with the current TSGV research and share our insights about promising research directions.

A Privacy-Preserving Unsupervised Domain Adaptation Framework for Clinical Text Analysis

Unsupervised domain adaptation (UDA) generally aligns the unlabeled target domain data to the distribution of the source domain to mitigate the distribution shift problem. The standard UDA requires sharing the source data with the target, having potential data privacy leaking risks. To protect the source data's privacy, we first propose to share the source feature distribution instead of the source data. However, sharing only the source feature distribution may still suffer from the membership inference attack who can infer an individual's membership by the black-box access to the source model. To resolve this privacy issue, we further study the under-explored problem of privacy-preserving domain adaptation and propose a method with a novel differential privacy training strategy to protect the source data privacy. We model the source feature distribution by Gaussian Mixture Models (GMMs) under the differential privacy setting and send it to the target client for adaptation. The target client resamples differentially private source features from GMMs and adapts on target data with several state-of-art UDA backbones. With our proposed method, the source data provider could avoid leaking source data privacy during domain adaptation as well as reserve the utility. To evaluate our proposed method's utility and privacy loss, we apply our model on a medical report disease label classification task using two noisy challenging clinical text datasets. The results show that our proposed method can preserve source data's privacy with a minor performance influence on the text classification task.

Neighborhood Region Smoothing Regularization for Finding Flat Minima In Deep Neural Networks

Due to diverse architectures in deep neural networks (DNNs) with severe overparameterization, regularization techniques are critical for finding optimal solutions in the huge hypothesis space. In this paper, we propose an effective regularization technique, called Neighborhood Region Smoothing (NRS). NRS leverages the finding that models would benefit from converging to flat minima, and tries to regularize the neighborhood region in weight space to yield approximate outputs. Specifically, gap between outputs of models in the neighborhood region is gauged by a defined metric based on Kullback-Leibler divergence. This metric provides similar insights with the minimum description length principle on interpreting flat minima. By minimizing both this divergence and empirical loss, NRS could explicitly drive the optimizer towards converging to flat minima. We confirm the effectiveness of NRS by performing image classification tasks across a wide range of model architectures on commonly-used datasets such as CIFAR and ImageNet, where generalization ability could be universally improved. Also, we empirically show that the minima found by NRS would have relatively smaller Hessian eigenvalues compared to the conventional method, which is considered as the evidence of flat minima.

Manifoldron: Direct Space Partition via Manifold Discovery

A neural network with the widely-used ReLU activation has been shown to partition the sample space into many convex polytopes for prediction. However, the parameterized way a neural network and other machine learning models use to partition the space has imperfections, e.g., the compromised interpretability for complex models, the inflexibility in decision boundary construction due to the generic character of the model, and the risk of being trapped into shortcut solutions. In contrast, although the non-parameterized models can adorably avoid or downplay these issues, they are usually insufficiently powerful either due to over-simplification or the failure to accommodate the manifold structures of data. In this context, we first propose a new type of machine learning models referred to as Manifoldron that directly derives decision boundaries from data and partitions the space via manifold structure discovery. Then, we systematically analyze the key characteristics of the Manifoldron including interpretability, manifold characterization capability, and its link to neural networks. The experimental results on 9 small and 11 large datasets demonstrate that the proposed Manifoldron performs competitively compared to the mainstream machine learning models. We have shared our code https://github.com/wdayang/Manifoldron for free download and evaluation.

SAC-GAN: Structure-Aware Image-to-Image Composition for Self-Driving

We present a compositional approach to image augmentation for self-driving applications. It is an end-to-end neural network that is trained to seamlessly compose an object (e.g., a vehicle or pedestrian) represented as a cropped patch from an object image, into a background scene image. As our approach emphasizes more on semantic and structural coherence of the composed images, rather than their pixel-level RGB accuracies, we tailor the input and output of our network with structure-aware features and design our network losses accordingly. Specifically, our network takes the semantic layout features from the input scene image, features encoded from the edges and silhouette in the input object patch, as well as a latent code as inputs, and generates a 2D spatial affine transform defining the translation and scaling of the object patch. The learned parameters are further fed into a differentiable spatial transformer network to transform the object patch into the target image, where our model is trained adversarially using an affine transform discriminator and a layout discriminator. We evaluate our network, coined SAC-GAN for structure-aware composition, on prominent self-driving datasets in terms of quality, composability, and generalizability of the composite images. Comparisons are made to state-of-the-art alternatives, confirming superiority of our method.

Face Deblurring Based on Separable Normalization and Adaptive Denormalization

Face deblurring aims to restore a clear face image from a blurred input image with more explicit structure and facial details. However, most conventional image and face deblurring methods focus on the whole generated image resolution without consideration of special face part texture and generally produce unsufficient details. Considering that faces and backgrounds have different distribution information, in this study, we designed an effective face deblurring network based on separable normalization and adaptive denormalization (SNADNet). First, We fine-tuned the face parsing network to obtain an accurate face structure. Then, we divided the face parsing feature into face foreground and background. Moreover, we constructed a new feature adaptive denormalization to regularize fafcial structures as a condition of the auxiliary to generate more harmonious and undistorted face structure. In addition, we proposed a texture extractor and multi-patch discriminator to enhance the generated facial texture information. Experimental results on both CelebA and CelebA-HQ datasets demonstrate that the proposed face deblurring network restores face structure with more facial details and performs favorably against state-of-the-art methods in terms of structured similarity indexing method (SSIM), peak signal-to-noise ratio (PSNR), Frechet inception distance (FID) and L1, and qualitative comparisons.

Hybrid Neural Networks for On-device Directional Hearing

On-device directional hearing requires audio source separation from a given direction while achieving stringent human-imperceptible latency requirements. While neural nets can achieve significantly better performance than traditional beamformers, all existing models fall short of supporting low-latency causal inference on computationally-constrained wearables. We present DeepBeam, a hybrid model that combines traditional beamformers with a custom lightweight neural net. The former reduces the computational burden of the latter and also improves its generalizability, while the latter is designed to further reduce the memory and computational overhead to enable real-time and low-latency operations. Our evaluation shows comparable performance to state-of-the-art causal inference models on synthetic data while achieving a 5x reduction of model size, 4x reduction of computation per second, 5x reduction in processing time and generalizing better to real hardware data. Further, our real-time hybrid model runs in 8 ms on mobile CPUs designed for low-power wearable devices and achieves an end-to-end latency of 17.5 ms.