Search Results for author: Jin Li

Found 71 papers, 21 papers with code

Cost-Effective Online Multi-LLM Selection with Versatile Reward Models

no code implementations26 May 2024 Xiangxiang Dai, Jin Li, Xutong Liu, Anqi Yu, John C. S. Lui

The NP-hard integer linear programming problem for selecting multiple LLMs with trade-off dilemmas is addressed by: i) decomposing the integer problem into a relaxed form by the local server, ii) utilizing a discretization rounding scheme that provides optimal LLM combinations by the scheduling cloud, and iii) continual online updates based on feedback.


Are You Copying My Prompt? Protecting the Copyright of Vision Prompt for VPaaS via Watermark

no code implementations24 May 2024 Huali Ren, Anli Yan, Chong-zhi Gao, Hongyang Yan, Zhenxin Zhang, Jin Li

Specifically, it utilizes a poison-only backdoor attack method to embed a watermark into the prompt and then employs a hypothesis-testing approach for remote verification of prompt ownership.

Backdoor Attack

ParamReL: Learning Parameter Space Representation via Progressively Encoding Bayesian Flow Networks

1 code implementation24 May 2024 Zhangkai Wu, Xuhui Fan, Jin Li, Zhilin Zhao, Hui Chen, Longbing Cao

Specifically, ParamReL proposes a \emph{self-}encoder to learn latent semantics directly from parameters, rather than from observations.

Disentanglement Learning Semantic Representations

Incremental Learning with Concept Drift Detection and Prototype-based Embeddings for Graph Stream Classification

no code implementations3 Apr 2024 Kleanthis Malialis, Jin Li, Christos G. Panayiotou, Marios M. Polycarpou

Data stream mining aims at extracting meaningful knowledge from continually evolving data streams, addressing the challenges posed by nonstationary environments, particularly, concept drift which refers to a change in the underlying data distribution over time.

Decision Making Incremental Learning

CATP: Cross-Attention Token Pruning for Accuracy Preserved Multimodal Model Inference

no code implementations2 Apr 2024 Ruqi Liao, Chuqing Zhao, Jin Li, Weiqi Feng

In response to the rising interest in large multimodal models, we introduce Cross-Attention Token Pruning (CATP), a precision-focused token pruning method.

Computational Efficiency

Open-Vocabulary Scene Text Recognition via Pseudo-Image Labeling and Margin Loss

no code implementations12 Mar 2024 Xuhua Ren, Hengcan Shi, Jin Li

In this paper, we propose a novel open-vocabulary text recognition framework, Pseudo-OCR, to recognize OOV words.

Image Inpainting Optical Character Recognition (OCR) +2

Understanding Missingness in Time-series Electronic Health Records for Individualized Representation

no code implementations24 Feb 2024 Ghadeer O. Ghosheh, Jin Li, Tingting Zhu

The lack of focus on missingness representation in an individualized way limits the full utilization of machine learning applications towards true personalization.

Time Series

UMG-CLIP: A Unified Multi-Granularity Vision Generalist for Open-World Understanding

no code implementations12 Jan 2024 Bowen Shi, Peisen Zhao, Zichen Wang, Yuhang Zhang, Yaoming Wang, Jin Li, Wenrui Dai, Junni Zou, Hongkai Xiong, Qi Tian, Xiaopeng Zhang

Vision-language foundation models, represented by Contrastive language-image pre-training (CLIP), have gained increasing attention for jointly understanding both vision and textual tasks.

Panoptic Segmentation Retrieval +1

IGNITE: Individualized GeNeration of Imputations in Time-series Electronic health records

1 code implementation9 Jan 2024 Ghadeer O. Ghosheh, Jin Li, Tingting Zhu

In IGNITE, we further propose a novel individualized missingness mask (IMM), which helps our model generate values based on the individual's observed data and missingness patterns.

Time Series

GanFinger: GAN-Based Fingerprint Generation for Deep Neural Network Ownership Verification

no code implementations25 Dec 2023 Huali Ren, Anli Yan, Xiaojun Ren, Pei-Gen Ye, Chong-zhi Gao, Zhili Zhou, Jin Li

To address these issues, we propose a network fingerprinting approach, named as GanFinger, to construct the network fingerprints based on the network behavior, which is characterized by network outputs of pairs of original examples and conferrable adversarial examples.

Curriculum-Enhanced Residual Soft An-Isotropic Normalization for Over-smoothness in Deep GNNs

1 code implementation13 Dec 2023 Jin Li, Qirong Zhang, Shuling Xu, Xinlong Chen, Longkun Guo, Yang-Geng Fu

Despite Graph neural networks' significant performance gain over many classic techniques in various graph-related downstream tasks, their successes are restricted in shallow models due to over-smoothness and the difficulties of optimizations among many other issues.

Node Classification

AiluRus: A Scalable ViT Framework for Dense Prediction

1 code implementation NeurIPS 2023 Jin Li, Yaoming Wang, Xiaopeng Zhang, Bowen Shi, Dongsheng Jiang, Chenglin Li, Wenrui Dai, Hongkai Xiong, Qi Tian

Specifically, at the intermediate layer of the ViT, we utilize a spatial-aware density-based clustering algorithm to select representative tokens from the token sequence.

object-detection Object Detection +1

Hybrid Distillation: Connecting Masked Autoencoders with Contrastive Learners

no code implementations28 Jun 2023 Bowen Shi, Xiaopeng Zhang, Yaoming Wang, Jin Li, Wenrui Dai, Junni Zou, Hongkai Xiong, Qi Tian

In order to better obtain both discrimination and diversity, we propose a simple but effective Hybrid Distillation strategy, which utilizes both the supervised/CL teacher and the MIM teacher to jointly guide the student model.

Contrastive Learning Representation Learning

G-STO: Sequential Main Shopping Intention Detection via Graph-Regularized Stochastic Transformer

no code implementations25 Jun 2023 Yuchen Zhuang, Xin Shen, Yan Zhao, Chaosheng Dong, Ming Wang, Jin Li, Chao Zhang

The detection of the underlying shopping intentions of users based on their historical interactions is a crucial aspect for e-commerce platforms, such as Amazon, to enhance the convenience and efficiency of their customers' shopping experiences.

Sequential Recommendation

Text Is All You Need: Learning Language Representations for Sequential Recommendation

1 code implementation23 May 2023 Jiacheng Li, Ming Wang, Jin Li, Jinmiao Fu, Xin Shen, Jingbo Shang, Julian McAuley

In this paper, we propose to model user preferences and item features as language representations that can be generalized to new items and datasets.

Representation Learning Sentence +1

Learning to Personalize Recommendation based on Customers' Shopping Intents

no code implementations9 May 2023 Xin Shen, Jiaying Shi, Sungro Yoon, Jon Katzur, Hanbo Wang, Jim Chan, Jin Li

In this work, we introduce Amazon's new system that explicitly identifies and utilizes each customer's high level shopping intents for personalizing recommendations.

Two-stage MR Image Segmentation Method for Brain Tumors based on Attention Mechanism

no code implementations17 Apr 2023 Li Zhu, Jiawei Jiang, Lin Lu, Jin Li

In response to this problem, we introduce the Coordinate Attention (CA) module to replace the Res Block to reduce the number of parameters, and cooperate with the spatial information extraction network above to strengthen the information extraction ability.

Brain Tumor Segmentation Generative Adversarial Network +3

Adapting Shortcut With Normalizing Flow: An Efficient Tuning Framework for Visual Recognition

1 code implementation CVPR 2023 Yaoming Wang, Bowen Shi, Xiaopeng Zhang, Jin Li, Yuchen Liu, Wenrui Dai, Chenglin Li, Hongkai Xiong, Qi Tian

To mitigate the computational and storage demands, recent research has explored Parameter-Efficient Fine-Tuning (PEFT), which focuses on tuning a minimal number of parameters for efficient adaptation.

Motif-aware temporal GCN for fraud detection in signed cryptocurrency trust networks

no code implementations22 Nov 2022 Song Li, Jiandong Zhou, Chong Mo, Jin Li, Geoffrey K. F. Tso, Yuxing Tian

Whereas in this study, we consider the evolving nature of cryptocurrency networks, and use local structural as well as the balance theory to guide the training process.

Fraud Detection

A Framework for Mutual Information-based MIMO Integrated Sensing and Communication Beamforming Design

no code implementations15 Nov 2022 Jin Li, Gui Zhou, Tantao Gong, Nan Liu

For the case of a single communication user, we consider three types of echo interference, no interference, a point interference, and an extended interference.

Integrated Sensing and Communication Beamforming Design Based on Mutual Information

no code implementations8 Nov 2022 Jin Li, Nan Liu

A closed-form solution with low complexity and a solution based on the semidefinite relaxation (SDR) method are provided to solve these two problems, respectively.

Learning from Students: Online Contrastive Distillation Network for General Continual Learning

1 code implementation Conference 2022 Jin Li, Zhong Ji, Gang Wang, Qiang Wang, Feng Gao

The goal of General Continual Learning (GCL) is to preserve learned knowledge and learn new knowledge with constant memory from an infinite data stream where task boundaries are blurry.

Continual Learning

Adaptive Domain Interest Network for Multi-domain Recommendation

no code implementations20 Jun 2022 Yuchen Jiang, Qi Li, Han Zhu, Jinbei Yu, Jin Li, Ziru Xu, Huihui Dong, Bo Zheng

Industrial recommender systems usually hold data from multiple business scenarios and are expected to provide recommendation services for these scenarios simultaneously.

Domain Adaptation Recommendation Systems +1

An efficient real-time target tracking algorithm using adaptive feature fusion

no code implementations5 Apr 2022 Yanyan Liu, Changcheng Pan, Minglin Bie, Jin Li

To address this issue, an efficient real-time target tracking method based on a low-dimension adaptive feature fusion is proposed to allow us the simultaneous implementation of the high-accuracy and real-time target tracking.

Dimensionality Reduction

Learning Personalized Item-to-Item Recommendation Metric via Implicit Feedback

no code implementations18 Mar 2022 Trong Nghia Hoang, Anoop Deoras, Tong Zhao, Jin Li, George Karypis

We develop and investigate a personalizable deep metric model that captures both the internal contents of items and how they were interacted with by users.

Metric Learning Recommendation Systems

A review of Generative Adversarial Networks for Electronic Health Records: applications, evaluation measures and data sources

no code implementations14 Mar 2022 Ghadeer Ghosheh, Jin Li, Tingting Zhu

Electronic Health Records (EHRs) are a valuable asset to facilitate clinical research and point of care applications; however, many challenges such as data privacy concerns impede its optimal utilization.

BIG-bench Machine Learning

AI-enabled Automatic Multimodal Fusion of Cone-Beam CT and Intraoral Scans for Intelligent 3D Tooth-Bone Reconstruction and Clinical Applications

no code implementations11 Mar 2022 Jin Hao, Jiaxiang Liu, Jin Li, Wei Pan, Ruizhe Chen, Huimin Xiong, Kaiwei Sun, Hangzheng Lin, Wanlu Liu, Wanghui Ding, Jianfei Yang, Haoji Hu, Yueling Zhang, Yang Feng, Zeyu Zhao, Huikai Wu, Youyi Zheng, Bing Fang, Zuozhu Liu, Zhihe Zhao

Here, we present a Deep Dental Multimodal Analysis (DDMA) framework consisting of a CBCT segmentation model, an intraoral scan (IOS) segmentation model (the most accurate digital dental model), and a fusion model to generate 3D fused crown-root-bone structures with high fidelity and accurate occlusal and dentition information.


PointSCNet: Point Cloud Structure and Correlation Learning Based on Space Filling Curve-Guided Sampling

1 code implementation21 Feb 2022 Xingye Chen, Yiqi Wu, Wenjie Xu, Jin Li, Huaiyi Dong, Yilin Chen

This paper proposes a point cloud feature extraction network named PointSCNet, to capture the geometrical structure information and local region correlation information of a point cloud.

3D Point Cloud Classification Semantic Segmentation

GraphEye: A Novel Solution for Detecting Vulnerable Functions Based on Graph Attention Network

no code implementations5 Feb 2022 Li Zhou, Minhuan Huang, YuJun Li, Yuanping Nie, Jin Li, Yiwei Liu

GraphEye is originated from the observation that the code property graph of a non-vulnerable function naturally differs from the code property graph of a vulnerable function with the same functionality.

C++ code Graph Attention +1

Learning Canonical F-Correlation Projection for Compact Multiview Representation

no code implementations CVPR 2022 Yun-Hao Yuan, Jin Li, Yun Li, Jipeng Qiang, Yi Zhu, Xiaobo Shen, Jianping Gou

With this framework as a tool, we propose a correlative covariation projection (CCP) method by using an explicit nonlinear mapping.

Representation Learning

Contrastive Regression for Domain Adaptation on Gaze Estimation

no code implementations CVPR 2022 Yaoming Wang, Yangzhou Jiang, Jin Li, Bingbing Ni, Wenrui Dai, Chenglin Li, Hongkai Xiong, Teng Li

Appearance-based Gaze Estimation leverages deep neural networks to regress the gaze direction from monocular images and achieve impressive performance.

Domain Generalization Gaze Estimation +1

Generating Synthetic Mixed-type Longitudinal Electronic Health Records for Artificial Intelligent Applications

1 code implementation22 Dec 2021 Jin Li, Benjamin J. Cairns, Jingsong Li, Tingting Zhu

Synthetic data, which benefits from the development and proliferation of generative models, has served as a promising substitute for real patient EHR data.

Decision Making Generative Adversarial Network

Neural Points: Point Cloud Representation with Neural Fields for Arbitrary Upsampling

1 code implementation CVPR 2022 Wanquan Feng, Jin Li, Hongrui Cai, Xiaonan Luo, Juyong Zhang

Different from traditional point cloud representation where each point only represents a position or a local plane in the 3D space, each point in Neural Points represents a local continuous geometric shape via neural fields.

point cloud upsampling

Enhanced Memory Network: The novel network structure for Symbolic Music Generation

1 code implementation7 Oct 2021 Jin Li, Haibin Liu, Nan Yan, Lan Wang

Symbolic melodies generation is one of the essential tasks for automatic music generation.

Music Generation

Understanding Self-supervised Learning via Information Bottleneck Principle

no code implementations29 Sep 2021 Jin Li, Yaoming Wang, Dongsheng Jiang, Xiaopeng Zhang, Wenrui Dai, Hongkai Xiong

To address this issue, we introduce the information bottleneck principle and propose the Self-supervised Variational Information Bottleneck (SVIB) learning framework.

Contrastive Learning Self-Supervised Learning

Complementary Calibration: Boosting General Continual Learning with Collaborative Distillation and Self-Supervision

1 code implementation3 Sep 2021 Zhong Ji, Jin Li, Qiang Wang, Zhongfei Zhang

Furthermore, we explore a collaborative self-supervision idea to leverage pretext tasks and supervised contrastive learning for addressing the feature deviation problem by learning complete and discriminative features for all classes.

Continual Learning Contrastive Learning +2

Dynamic Selection in Algorithmic Decision-making

no code implementations28 Aug 2021 Jin Li, Ye Luo, Xiaowei Zhang

This paper identifies and addresses dynamic selection problems in online learning algorithms with endogenous data.

Decision Making

Unsupervised Cross-Lingual Speech Emotion Recognition Using Pseudo Multilabel

1 code implementation19 Aug 2021 Jin Li, Nan Yan, Lan Wang

However, cross-lingual SER remains a challenge in real-world applications due to a great difference between the source and target domain distributions.

Speech Emotion Recognition

FDN: Finite Difference Network with Hierarchical Convolutional Features for Text-independent Speaker Verification

1 code implementation18 Aug 2021 Jin Li, Nan Yan, Lan Wang

For example, RawNet and RawNet2 extracted speaker's feature embeddings from waveforms automatically for recognizing their voice, which can vastly reduce the front-end computation and obtain state-of-the-art performance.

Text-Independent Speaker Verification

A Multi-level Acoustic Feature Extraction Framework for Transformer Based End-to-End Speech Recognition

no code implementations18 Aug 2021 Jin Li, Rongfeng Su, Xurong Xie, Nan Yan, Lan Wang

The shallow stream is used to acquire traditional shallow features that is beneficial for the classification of phones or words while the deep stream is used to obtain utterance-level speaker-invariant deep features for improving the feature diversity.

Automatic Speech Recognition Automatic Speech Recognition (ASR) +3

Wavelet Transform-assisted Adaptive Generative Modeling for Colorization

4 code implementations9 Jul 2021 Jin Li, Wanyun Li, Zichen Xu, Yuhao Wang, Qiegen Liu

Unsupervised deep learning has recently demonstrated the promise of producing high-quality samples.

Colorization Denoising +1

Improving Cost Learning for JPEG Steganography by Exploiting JPEG Domain Knowledge

no code implementations9 May 2021 Weixuan Tang, Bin Li, Mauro Barni, Jin Li, Jiwu Huang

To address the issue, in this paper we extend an existing automatic cost learning scheme to JPEG, where the proposed scheme called JEC-RL (JPEG Embedding Cost with Reinforcement Learning) is explicitly designed to tailor the JPEG DCT structure.

reinforcement-learning Reinforcement Learning (RL)

Deep Music Retrieval for Fine-Grained Videos by Exploiting Cross-Modal-Encoded Voice-Overs

1 code implementation21 Apr 2021 Tingtian Li, Zixun Sun, Haoruo Zhang, Jin Li, Ziming Wu, Hui Zhan, Yipeng Yu, Hengcan Shi

In this paper, we also investigate the widely added voice-overs in short videos and propose a novel framework to retrieve BGM for fine-grained short videos.

Pseudo Label Retrieval

Causal Reinforcement Learning: An Instrumental Variable Approach

no code implementations6 Mar 2021 Jin Li, Ye Luo, Xiaowei Zhang

In the standard data analysis framework, data is first collected (once for all), and then data analysis is carried out.

Causal Inference reinforcement-learning +1

Truncation-Free Matching System for Display Advertising at Alibaba

no code implementations18 Feb 2021 Jin Li, Jie Liu, Shangzhou Li, Yao Xu, Ran Cao, Qi Li, Biye Jiang, Guan Wang, Han Zhu, Kun Gai, Xiaoqiang Zhu

When receiving a user request, matching system (i) finds the crowds that the user belongs to; (ii) retrieves all ads that have targeted those crowds.


Simulation on the Transparency of Electrons and Ion Back Flow for a Time Projection Chamber based on Staggered Multiple THGEMs

no code implementations16 Feb 2021 Mengzhi Wu, Qian Liu, Ping Li, Shi Chen, Binlong Wang, Wenhan Shen, Shiping Chen, Yangheng Zheng, Yigang Xie, Jin Li

The IBF and the transparent rate of electrons are two essential indicators of TPC, which affect the energy resolution and counting rate respectively.

Instrumentation and Detectors High Energy Physics - Experiment

Joint Intensity-Gradient Guided Generative Modeling for Colorization

7 code implementations28 Dec 2020 Kai Hong, Jin Li, Wanyun Li, Cailian Yang, Minghui Zhang, Yuhao Wang, Qiegen Liu

Furthermore, the joint intensity-gradient constraint in data-fidelity term is proposed to limit the degree of freedom within generative model at the iterative colorization stage, and it is conducive to edge-preserving.


A Reversed and Shift Sparse Array Scheme based on the Difference and Sum Co-array

no code implementations26 Nov 2020 Yan Zhou, Jin Li, Nieke Wei

The reversed and shift (RAS) sparse array scheme, which is based on the difference and sum co-array (DSCA) and remarkably enhances the capability of identifying sources, is proposed.

An Automatic Cost Learning Framework for Image Steganography Using Deep Reinforcement Learning

1 code implementation journal 2020 Weixuan Tang, Bin Li, Mauro Barni, Jin Li, Jiwu Huang

In SPAR-RL, an agent utilizes a policy network which decomposes the embedding process into pixel-wise actions and aims at maximizing the total rewards from a simulated steganalytic environment, while the environment employs an environment network for pixel-wise reward assignment.

Image Steganography reinforcement-learning

A Deep Prediction Network for Understanding Advertiser Intent and Satisfaction

no code implementations20 Aug 2020 Liyi Guo, Rui Lu, Haoqi Zhang, Junqi Jin, Zhenzhe Zheng, Fan Wu, Jin Li, Haiyang Xu, Han Li, Wenkai Lu, Jian Xu, Kun Gai

For e-commerce platforms such as Taobao and Amazon, advertisers play an important role in the entire digital ecosystem: their behaviors explicitly influence users' browsing and shopping experience; more importantly, advertiser's expenditure on advertising constitutes a primary source of platform revenue.


Publicly Verifiable Databases With All Efficient Updating Operations

no code implementations IEEE Transactions on Knowledge and Data Engineering 2020 Xiaofeng Chen, Hui Li, Jin Li, Qian Wang, Xinyi Huang, Willy Susilo, and Yang Xiang

As a result, it remains an open problem how to construct an efficient (and publicly verifiable) VDB scheme that can support all updating operations regardless of the manner of insertion.

Fast Reinforcement Learning for Anti-jamming Communications

no code implementations13 Feb 2020 Pei-Gen Ye, Yuan-Gen Wang, Jin Li, Liang Xiao

This letter presents a fast reinforcement learning algorithm for anti-jamming communications which chooses previous action with probability $\tau$ and applies $\epsilon$-greedy with probability $(1-\tau)$.

reinforcement-learning Reinforcement Learning (RL)

Controllable List-wise Ranking for Universal No-reference Image Quality Assessment

1 code implementation24 Nov 2019 Fu-Zhao Ou, Yuan-Gen Wang, Jin Li, Guopu Zhu, Sam Kwong

First, to extend the authentically distorted image dataset, we present an imaging-heuristic approach, in which the over-underexposure is formulated as an inverse of Weber-Fechner law, and fusion strategy and probabilistic compression are adopted, to generate the degraded real-world images.

No-Reference Image Quality Assessment NR-IQA

DASGrad: Double Adaptive Stochastic Gradient

no code implementations25 Sep 2019 Kin Gutierrez, Cristian Challu, Jin Li, Artur Dubrawski

Adaptive moment methods have been remarkably successful for optimization under the presence of high dimensional or sparse gradients, in parallel to this, adaptive sampling probabilities for SGD have allowed optimizers to improve convergence rates by prioritizing examples to learn efficiently.

Transfer Learning

Universal Transforming Geometric Network

1 code implementation2 Aug 2019 Jin Li

The recurrent geometric network (RGN), the first end-to-end differentiable neural architecture for protein structure prediction, is a competitive alternative to existing models.

Protein Structure Prediction

Compressing Unknown Images With Product Quantizer for Efficient Zero-Shot Classification

no code implementations CVPR 2019 Jin Li, Xuguang Lan, Yang Liu, Le Wang, Nanning Zheng

Based on this intuition, a Product Quantization Zero-Shot Learning (PQZSL) method is proposed to learn embeddings as well as quantizers to compress visual features into compact codes for Approximate NN (ANN) search.

General Classification Generalized Zero-Shot Learning +1

Double Adaptive Stochastic Gradient Optimization

no code implementations6 Nov 2018 Kin Gutierrez, Jin Li, Cristian Challu, Artur Dubrawski

We observe that the benefits of~\textsc{DASGrad} increase with the model complexity and variability of the gradients, and we explore the resulting utility in extensions of distribution-matching multitask learning.

Security Matters: A Survey on Adversarial Machine Learning

no code implementations16 Oct 2018 Guofu Li, Pengjia Zhu, Jin Li, Zhemin Yang, Ning Cao, Zhiyi Chen

Adversarial machine learning is a fast growing research area, which considers the scenarios when machine learning systems may face potential adversarial attackers, who intentionally synthesize input data to make a well-trained model to make mistake.

BIG-bench Machine Learning

A Real-time Robotic Grasp Approach with Oriented Anchor Box

no code implementations8 Sep 2018 Hanbo Zhang, Xinwen Zhou, Xuguang Lan, Jin Li, Zhiqiang Tian, Nanning Zheng

The main component of our approach is a grasp detection network with oriented anchor boxes as detection priors.


Zero-Shot Learning by Generating Pseudo Feature Representations

no code implementations19 Mar 2017 Jiang Lu, Jin Li, Ziang Yan, Chang-Shui Zhang

Given the dataset of seen classes and side information of unseen classes (e. g. attributes), we synthesize feature-level pseudo representations for novel concepts, which allows us access to the formulation of unseen class predictor.

Attribute Novel Concepts +2

Weakly- and Semi-Supervised Object Detection with Expectation-Maximization Algorithm

no code implementations28 Feb 2017 Ziang Yan, Jian Liang, Weishen Pan, Jin Li, Chang-Shui Zhang

Object detection when provided image-level labels instead of instance-level labels (i. e., bounding boxes) during training is an important problem in computer vision, since large scale image datasets with instance-level labels are extremely costly to obtain.

object-detection Object Detection +1

Cannot find the paper you are looking for? You can Submit a new open access paper.