在本文中，我们通过变异自动编码器（VAE）研究了基于弦的分子生成的问题，这些问题已经为人工智能的各种任务提供了一种流行的生成方法。我们提出了一个简单而有效的想法，以提高VAE的任务绩效。我们的主要思想是在共享单个编码器时维护多个解码器，即它是一种合奏技术。在这里，我们首先发现，由于合奏解码器的偏见在其自动回归推理下严重增加，因此每个解码器都可能没有有效。为了维持集合模型的较小偏见和差异，我们提出的技术是两倍：（a）为每个解码器采样不同的潜在变量（从共享编码器提供的估计平均值和差异）来鼓励解码器的多元化特征（b）在培训期间使用协作损失，以控制使用不同的潜在变量的解码器的汇总质量。在我们的实验中，提出的VAE模型特别表现出色，可从域外分布产生样品。

大多数物体检测方法通过使用非最大抑制（NMS）及其改进版本，如Soft-NMS获取对象，这是一个很长的历史记录，以删除冗余边界框。我们从三个方面挑战那些基于NMS的方法：1）具有最高置信度值的边界框可能不是具有与地面真理盒最大的重叠的真正积极。 2）冗余盒不仅需要抑制，而且对于那些真正的阳性也需要置信度。 3）不需要置信度值排序候选盒，以便可以实现完整的并行性。在本文中，通过信仰传播（BP）的启发，我们提出了置信沟集团（CP簇）来替换基于NMS的方法，这是完全并行化的，以及精度更好。在CP-Cluster中，我们借用BP的消息传递机制来惩罚冗余框，并以迭代方式同时增强真正的阳性直到收敛。我们通过将其应用于各种主流探测器，例如FasterRCNN，SSD，FCO，YOLOV3，YOLOV5，CENTERENET等实验，验证了CP-Cluster的有效性。在MS COCO上的实验表明，我们的插头和游戏方法没有再培训探测器，都能够稳步与基于NMS的方法相比，将分别从0.2到1.9的透明边距提高所有最先进模型的平均地图。源代码在https://github.com/shenyi0220/cp-cluster中获得

Several leading methods on public benchmarks for depth-from-stereo rely on memory-demanding 4D cost volumes and computationally intensive 3D convolutions for feature matching. We suggest a new way to process the 4D cost volume where we merge two different concepts in one deeply integrated framework to achieve a symbiotic relationship. A feature matching part is responsible for identifying matching pixels pairs along the baseline while a concurrent image volume part is inspired by depth-from-mono CNNs. However, instead of predicting depth directly from image features, it provides additional context to resolve ambiguities during pixel matching. More technically, the processing of the 4D cost volume is separated into a 2D propagation and a 3D propagation part. Starting from feature maps of the left image, the 2D propagation assists the 3D propagation part of the cost volume at different layers by adding visual features to the geometric context. By combining both parts, we can safely reduce the scale of 3D convolution layers in the matching part without sacrificing accuracy. Experiments demonstrate that our end-to-end trained CNN is ranked 2nd on KITTI2012 and ETH3D benchmarks while being significantly faster than the 1st-ranked method. Furthermore, we notice that the coupling of image and matching-volume improves fine-scale details as demonstrated by our qualitative analysis.

Uniform-precision neural network quantization has gained popularity since it simplifies densely packed arithmetic unit for high computing capability. However, it ignores heterogeneous sensitivity to the impact of quantization errors across the layers, resulting in sub-optimal inference accuracy. This work proposes a novel neural architecture search called neural channel expansion that adjusts the network structure to alleviate accuracy degradation from ultra-low uniform-precision quantization. The proposed method selectively expands channels for the quantization sensitive layers while satisfying hardware constraints (e.g., FLOPs, PARAMs). Based on in-depth analysis and experiments, we demonstrate that the proposed method can adapt several popular networks channels to achieve superior 2-bit quantization accuracy on CIFAR10 and ImageNet. In particular, we achieve the best-to-date Top-1/Top-5 accuracy for 2-bit ResNet50 with smaller FLOPs and the parameter size.

Many real-world applications of language models (LMs), such as code autocomplete and writing assistance, involve human-LM interaction, but the main LM benchmarks are non-interactive, where a system produces output without human intervention. To evaluate human-LM interaction, we develop a framework, Human-AI Language-based Interaction Evaluation (H-LINE), that expands non-interactive evaluation along three dimensions, capturing (i) the interactive process, not only the final output; (ii) the first-person subjective experience, not just a third-party assessment; and (iii) notions of preference beyond quality. We then design five tasks ranging from goal-oriented to open-ended to capture different forms of interaction. On four state-of-the-art LMs (three variants of OpenAI's GPT-3 and AI21's J1-Jumbo), we find that non-interactive performance does not always result in better human-LM interaction and that first-person and third-party metrics can diverge, suggesting the importance of examining the nuances of human-LM interaction.

Cartoonization is a task that renders natural photos into cartoon styles. Previous deep cartoonization methods only have focused on end-to-end translation, which may hinder editability. Instead, we propose a novel solution with editing features of texture and color based on the cartoon creation process. To do that, we design a model architecture to have separate decoders, texture and color, to decouple these attributes. In the texture decoder, we propose a texture controller, which enables a user to control stroke style and abstraction to generate diverse cartoon textures. We also introduce an HSV color augmentation to induce the networks to generate diverse and controllable color translation. To the best of our knowledge, our work is the first deep approach to control the cartoonization at inference while showing profound quality improvement over to baselines.

Brain-computer interface (BCI) uses brain signals to communicate with external devices without actual control. Particularly, BCI is one of the interfaces for controlling the robotic arm. In this study, we propose a knowledge distillation-based framework to manipulate robotic arm through hybrid paradigm induced EEG signals for practical use. The teacher model is designed to decode input data hierarchically and transfer knowledge to student model. To this end, soft labels and distillation loss functions are applied to the student model training. According to experimental results, student model achieved the best performance among the singular architecture-based methods. It is confirmed that using hierarchical models and knowledge distillation, the performance of a simple architecture can be improved. Since it is uncertain what knowledge is transferred, it is important to clarify this part in future studies.

Score-based generative models are shown to achieve remarkable empirical performances in various applications such as image generation and audio synthesis. However, a theoretical understanding of score-based diffusion models is still incomplete. Recently, Song et al. showed that the training objective of score-based generative models is equivalent to minimizing the Kullback-Leibler divergence of the generated distribution from the data distribution. In this work, we show that score-based models also minimize the Wasserstein distance between them under suitable assumptions on the model. Specifically, we prove that the Wasserstein distance is upper bounded by the square root of the objective function up to multiplicative constants and a fixed constant offset. Our proof is based on a novel application of the theory of optimal transport, which can be of independent interest to the society. Our numerical experiments support our findings. By analyzing our upper bounds, we provide a few techniques to obtain tighter upper bounds.

Emerging real-time multi-model ML (RTMM) workloads such as AR/VR and drone control often involve dynamic behaviors in various levels; task, model, and layers (or, ML operators) within a model. Such dynamic behaviors are new challenges to the system software in an ML system because the overall system load is unpredictable unlike traditional ML workloads. Also, the real-time processing requires to meet deadlines, and multi-model workloads involve highly heterogeneous models. As RTMM workloads often run on resource-constrained devices (e.g., VR headset), developing an effective scheduler is an important research problem. Therefore, we propose a new scheduler, SDRM3, that effectively handles various dynamicity in RTMM style workloads targeting multi-accelerator systems. To make scheduling decisions, SDRM3 quantifies the unique requirements for RTMM workloads and utilizes the quantified scores to drive scheduling decisions, considering the current system load and other inference jobs on different models and input frames. SDRM3 has tunable parameters that provide fast adaptivity to dynamic workload changes based on a gradient descent-like online optimization, which typically converges within five steps for new workloads. In addition, we also propose a method to exploit model level dynamicity based on Supernet for exploiting the trade-off between the scheduling effectiveness and model performance (e.g., accuracy), which dynamically selects a proper sub-network in a Supernet based on the system loads. In our evaluation on five realistic RTMM workload scenarios, SDRM3 reduces the overall UXCost, which is a energy-delay-product (EDP)-equivalent metric for real-time applications defined in the paper, by 37.7% and 53.2% on geometric mean (up to 97.6% and 97.1%) compared to state-of-the-art baselines, which shows the efficacy of our scheduling methodology.

Vision Transformers (ViTs) have become a dominant paradigm for visual representation learning with self-attention operators. Although these operators provide flexibility to the model with their adjustable attention kernels, they suffer from inherent limitations: (1) the attention kernel is not discriminative enough, resulting in high redundancy of the ViT layers, and (2) the complexity in computation and memory is quadratic in the sequence length. In this paper, we propose a novel attention operator, called lightweight structure-aware attention (LiSA), which has a better representation power with log-linear complexity. Our operator learns structural patterns by using a set of relative position embeddings (RPEs). To achieve log-linear complexity, the RPEs are approximated with fast Fourier transforms. Our experiments and ablation studies demonstrate that ViTs based on the proposed operator outperform self-attention and other existing operators, achieving state-of-the-art results on ImageNet, and competitive results on other visual understanding benchmarks such as COCO and Something-Something-V2. The source code of our approach will be released online.