Abstract Scalable, high-capacity, and low-power computing architecture is the primary assurance for increasingly manifold large-scale machine learning tasks. Traditional electronic artificial agents by conventional power-hungry processors have faced issues of energy scaling walls, hindering them from sustainable performance improvement iterative multi-task learning. Referring to another modality light, photonic has been progressively applied in high-efficient neuromorphic systems. Here, we...

Kun Zhou, Xiaolei Wang, Yuanhang Chenzhan Shang, Yuan Cheng, Wayne Xin Zhao, Yaliang Li, Ji-Rong Wen. Proceedings of the 59th Annual Meeting Association for Computational Linguistics and 11th International Joint Conference on Natural Language Processing: System Demonstrations. 2021.

Magnetic tunneling junctions (MTJs) lie in the core of magnetic random access memory, holding promise integrating memory and computing to reduce hardware complexity, transition latency, power consumption. However, traditional MTJs are insensitive light, limiting their functionality in-memory sensing─a crucial component for machine vision systems artificial intelligence applications. Herein, convergence with optical sensing capabilities is achieved all-two-dimensional (2D) junction Fe

Positron Emission Tomography (PET) imaging plays a crucial role in modern medical diagnostics by revealing the metabolic processes within patient's body, which is essential for quantification of therapy response and monitoring treatment progress. However, segmentation PET images presents unique challenges due to their lower contrast less distinct boundaries compared other structural modalities. Recent developments foundation models have shown superior versatility across diverse natural image...

With the explosive growth of web videos in recent years, large-scale Content-Based Video Retrieval (CBVR) becomes increasingly essential video filtering, recommendation, and copyright protection. Segment-level CBVR (S-CBVR) locates start end time similar segments finer granularity, which is beneficial for user browsing efficiency infringement detection especially long scenarios. The challenge S-CBVR task how to achieve high temporal alignment accuracy with efficient computation low storage...

In this paper, we introduce VCSL (Video Copy Segment Localization), a new comprehensive segment-level annotated video copy dataset. Compared with existing detection datasets restricted by either video-level annotation or small-scale, not only has two orders of magnitude more labelled data, 160k realistic pairs containing than 280k localized copied segment pairs, but also covers variety categories and wide range duration. All the segments inside each collected pair are manually extracted...

In recent years, the explosion of web videos makes text-video retrieval increasingly essential and popular for video filtering, recommendation, search. Text-video aims to rank relevant text/video higher than irrelevant ones. The core this task is precisely measure cross-modal similarity between texts videos. Recently, contrastive learning methods have shown promising results retrieval, most which focus on construction positive negative pairs learn text representations. Nevertheless, they do...

Recently, with the emergence of retrieval requirements for certain individual in same superclass, e.g., birds, persons, cars, fine-grained recognition task has attracted a significant amount attention from academia and industry. In scenario, inter-class differences are quite diverse subtle, which makes it challenging to extract all discriminative cues. Traditional training mechanism optimizes overall discriminativeness whole feature. It may stop early when some feature elements been trained...

Influencer marketing is emerging as a new method, changing the strategies of brands profoundly. In order to help find suitable micro-influencers partners, micro-influencer recommendation regarded an indispensable part influencer marketing. However, previous works only focus on modeling <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">individual image</i> brands/micro-influencers, which insufficient represent characteristics...

Product image search in E-commerce systems is a challenging task, because of huge number product classes, low intra-class similarity and high inter-class similarity. Deep metric learning, based on paired distances independent the aims to minimize variances feature embedding space. Most existing approaches strictly restrict distance between samples with fixed values distinguish different classes samples. However, has various magnitudes during training stages. Therefore, it difficult directly...

We illustrate how one can use basic combinatorial theory and computer programming technique (Python) to analyze the game: Mahjong. The results confirm some folklore concerning game, expose unexpected results. Related possible future research in connection artificial intelligence are mentioned. Readers interested subject may further develop techniques deepen study of or other games.

As it requires a huge number of parameters when exposed to high dimensional inputs in video detection and classification, there is grand challenge develop compact yet accurate comprehension at terminal devices. Current works focus on optimizations classification separated fashion. In this paper, we introduce (object action recognition) system for devices, namely DEEPEYE. Based You Only Look Once (YOLO), have developed an 8-bit quantization method training YOLO; also tensorized-compression...

Uncertain data may exist in many application fields, due to the inaccurate raw data, use of coarse-grained set, for purposes privacy protection, and integration etc. The original features be changed or ignored if uncertainties were mishandled. Therefore effective management analysis uncertain objects should rely on an appropriate model depicting characteristic uncertainties. For values attributes, this paper proposed construction method based nonparametric estimation, which can represent...

A novel method was brought forward for the purpose of filtering Gaussian noise effectively by using variable step time matrix simplified pulse coupled neural network (PCNN). Firstly, PCNN, related to grayscale and spatial information an image, is calculated identify polluted pixels. Subsequently, a step, long strong short weak noise, based on applied modify noised pixels in sliding window. And then wiener filter used image further noise. Experiments show that proposed can remove than other...

As the aging population continues to grow, fall detection has become a key issue in public health and healthcare. To address problem of low accuracy poor real-time performance algorithms real scenarios, an improved model, FALLNET, is proposed. First, YOLOv7-X-pose algorithm used quickly extract multiple human body keypoints multi-person keypoint extraction module. Second, classic CNN_Attention_LSTM model dangerous action recognition module by adding LSTM layer better capture important...

Recently, significant advancements have been made in Large Language Models (LLMs) through the implementation of various alignment techniques. These techniques enable LLMs to generate highly tailored content response diverse user instructions. Consequently, potential serve as robust, customizable recommendation systems field recommendation. However, using with individual information and online exploration remains a challenge, which are important perspectives developing personalized news...

Abstract. Stone cultural heritage, encompassing a broad spectrum of artifacts such as stone artworks, buildings, tools, and utensils, represents one the most significant categories heritage. However, conservation these heritage faces challenges from process deterioration. This degradation not only compromises structural integrity but also results in loss invaluable historical information. Thus, there emerges critical demand for effective methods to detect assess condition enabling timely...

Effective collaboration in multi-agent systems requires communicating goals and intentions between agents. Current agent frameworks often suffer from dependencies on single-agent execution lack robust inter-module communication, frequently leading to suboptimal reinforcement learning (MARL) policies inadequate task coordination. To address these challenges, we present a framework for training large language models (LLMs) as collaborative agents enable coordinated behaviors cooperative MARL....

Stent migration is one of the common complications after tracheal stent implantation. The causes include size mismatch between and trachea, physiological movement so on. In order to solve above problems, this study designed a non-uniform Poisson ratio by combining structure trachea improve stent, meanwhile ensuring support stent. study, corresponding cartilage was constructed with negative Poisson's ratio, circular connective tissue muscular membrane positive ratio. And four kinds stents...

In recent years, conversational recommender system (CRS) has received much attention in the research community. However, existing studies on CRS vary scenarios, goals and techniques, lacking unified, standardized implementation or comparison. To tackle this challenge, we propose an open-source toolkit CRSLab, which provides a unified extensible framework with highly-decoupled modules to develop CRSs. Based framework, collect 6 commonly-used human-annotated datasets implement 18 models that...