Searcharxiv⌕ Search

arXiv subjects

Chen Huang

Publications and source records attributed to Chen Huang.

At least 127 records · Page 7Linked to original sources
arXiv2022

Open-Vocabulary DETR with Conditional Matching

Open-vocabulary object detection, which is concerned with the problem of detecting novel objects guided by natural language, has gained increasing attention from the community. Ideally, we would like to extend an open-vocabulary detector such that it can produce bounding box predictions based on user inputs in form of either natural language or exemplar image. This offers great flexibility and user experience for human-computer interaction. To this end, we propose a novel open-vocabulary detector based on DETR -- hence the name OV-DETR -- which, once trained, can detect any object given its class name or an exemplar image. The biggest challenge of turning DETR into an open-vocabulary detector is that it is impossible to calculate the classification cost matrix of novel classes without access to their labeled images. To overcome this challenge, we formulate the learning objective as a binary matching one between input queries (class name or exemplar image) and the corresponding objects, which learns useful correspondence to generalize to unseen queries during testing. For training, we choose to condition the Transformer decoder on the input embeddings obtained from a pre-trained vision-language model like CLIP, in order to enable matching for both text and image queries. With extensive experiments on LVIS and COCO datasets, we demonstrate that our OV-DETR -- the first end-to-end Transformer-based open-vocabulary detector -- achieves non-trivial improvements over current state of the arts.

cs.CV↗
arXiv2022

Unified Vision and Language Prompt Learning

Prompt tuning, a parameter- and data-efficient transfer learning paradigm that tunes only a small number of parameters in a model's input space, has become a trend in the vision community since the emergence of large vision-language models like CLIP. We present a systematic study on two representative prompt tuning methods, namely text prompt tuning and visual prompt tuning. A major finding is that none of the unimodal prompt tuning methods performs consistently well: text prompt tuning fails on data with high intra-class visual variances while visual prompt tuning cannot handle low inter-class variances. To combine the best from both worlds, we propose a simple approach called Unified Prompt Tuning (UPT), which essentially learns a tiny neural network to jointly optimize prompts across different modalities. Extensive experiments on over 11 vision datasets show that UPT achieves a better trade-off than the unimodal counterparts on few-shot learning benchmarks, as well as on domain generalization benchmarks. Code and models will be released to facilitate future research.

cs.CV↗
arXiv2022

A Weighted Random Forest Based PositioningAlgorithm for 6G Indoor Communications

Due to the indoor none-line-of-sight (NLoS) propagation and multi-access interference (MAI), it is a great challenge to achieve centimeter-level positioning accuracy in indoor scenarios. However, the sixth generation (6G) wireless communications provide a good opportunity for the centimeter-level positioning. In 6G, the millimeter wave (mmWave) and terahertz (THz) communications have ultra-broad bandwidth so that the channel state information (CSI) will have a high resolution. In this paper, a weighted random forest (WRF) based indoor positioning algorithm using CSI based channel fingerprint feature is proposed to achieve high-precision positioning for 6G indoor communications. In addition, ray-tracing (RT) is used to improve the efficiency of establishing channel fingerprint database. The simulation results demonstrate the accuracy and robustness of the proposed algorithm. It is shown that the positioning accuracy of the algorithm is stable within 6 cm in different indoor scenarios with the channel fingerprint database established at 0.2 m intervals.

eess.SP↗
arXiv2022

An Empirical Study of Language Model Integration for Transducer based Speech Recognition

Utilizing text-only data with an external language model (ELM) in end-to-end RNN-Transducer (RNN-T) for speech recognition is challenging. Recently, a class of methods such as density ratio (DR) and internal language model estimation (ILME) have been developed, outperforming the classic shallow fusion (SF) method. The basic idea behind these methods is that RNN-T posterior should first subtract the implicitly learned internal language model (ILM) prior, in order to integrate the ELM. While recent studies suggest that RNN-T only learns some low-order language model information, the DR method uses a well-trained neural language model with full context, which may be inappropriate for the estimation of ILM and deteriorate the integration performance. Based on the DR method, we propose a low-order density ratio method (LODR) by replacing the estimation with a low-order weak language model. Extensive empirical experiments are conducted on both in-domain and cross-domain scenarios on English LibriSpeech & Tedlium-2 and Chinese WenetSpeech & AISHELL-1 datasets. It is shown that LODR consistently outperforms SF in all tasks, while performing generally close to ILME and better than DR in most tests.

eess.AS↗
arXiv2022

Position Prediction as an Effective Pretraining Strategy

Transformers have gained increasing popularity in a wide range of applications, including Natural Language Processing (NLP), Computer Vision and Speech Recognition, because of their powerful representational capacity. However, harnessing this representational capacity effectively requires a large amount of data, strong regularization, or both, to mitigate overfitting. Recently, the power of the Transformer has been unlocked by self-supervised pretraining strategies based on masked autoencoders which rely on reconstructing masked inputs, directly, or contrastively from unmasked content. This pretraining strategy which has been used in BERT models in NLP, Wav2Vec models in Speech and, recently, in MAE models in Vision, forces the model to learn about relationships between the content in different parts of the input using autoencoding related objectives. In this paper, we propose a novel, but surprisingly simple alternative to content reconstruction~-- that of predicting locations from content, without providing positional information for it. Doing so requires the Transformer to understand the positional relationships between different parts of the input, from their content alone. This amounts to an efficient implementation where the pretext task is a classification problem among all possible positions for each input token. We experiment on both Vision and Speech benchmarks, where our approach brings improvements over strong supervised training baselines and is comparable to modern unsupervised/self-supervised pretraining methods. Our method also enables Transformers trained without position embeddings to outperform ones trained with full position information.

cs.LG↗
arXiv2022

Efficient Representation Learning via Adaptive Context Pooling

Self-attention mechanisms model long-range context by using pairwise attention between all input tokens. In doing so, they assume a fixed attention granularity defined by the individual tokens (e.g., text characters or image pixels), which may not be optimal for modeling complex dependencies at higher levels. In this paper, we propose ContextPool to address this problem by adapting the attention granularity for each token. Inspired by the success of ConvNets that are combined with pooling to capture long-range dependencies, we learn to pool neighboring features for each token before computing attention in a given attention layer. The pooling weights and support size are adaptively determined, allowing the pooled features to encode meaningful context with varying scale. We show that ContextPool makes attention models more expressive, achieving strong performance often with fewer layers and thus significantly reduced cost. Experiments validate that our ContextPool module, when plugged into transformer models, matches or surpasses state-of-the-art performance using less compute on several language and image benchmarks, outperforms recent works with learned context sizes or sparse attention patterns, and is also applicable to ConvNets for efficient feature learning.

cs.LG↗
arXiv2022

Evading Thermodynamic Uncertainty Relations via Asymmetric Dynamic Protocols

Many versions of Thermodynamic Uncertainty Relations (TUR) have recently been discovered, which impose lower bounds on relative fluctuations of integrated currents in irreversible dissipative processes, and suggest that there may be fundamental limitations on the precision of small scale machines and heat engines. In this work we rigorously demonstrate that TUR can be evaded by using dynamic protocols that are asymmetric under time-reversal. We illustrate our results using a model heat engine using two-level systems, and also discuss heuristically the fundamental connections between TUR and time-reversal symmetry.

cond-mat.stat-mech↗
arXiv2021

A Scheme of Channel Prediction Based on Artificial Neural Network

Accurate channel modeling is the foundation of communication system design. However, the traditional measurement-based modeling approach has increasing challenges for the scenarios with insufficient measurement data. To obtain enough data for channel modeling, the Artificial Neural Network (ANN) is used in this paper to predict channel data. The high mobility railway channel is considered, which is a typical scenario where it is challenging to obtain enough data for modeling within a short sampling interval. Three types of ANNs, the Back Propagation Network, Radial Basis Function Neural Network and Extreme Learning Machine, are considered to predict channel path loss and shadow fading. The Root-Mean-Square error is used to evaluate prediction accuracy. The factors that may influence prediction accuracy are compared and discussed, including the type of network, number of neurons and proportion of training data. It is found that a larger number of neurons can significantly reduce prediction error, whereas the influence of proportion of training data is relatively small. The results can be used to improve modeling accuracy of path loss and shadow fading when measurement data is reduced.

cs.IT↗
arXiv2021

Artificial intelligence enabled radio propagation for communications-Part I: Channel characterization and antenna-channel optimization

To provide higher data rates, as well as better coverage, cost efficiency, security, adaptability, and scalability, the 5G and beyond 5G networks are developed with various artificial intelligence techniques. In this two-part paper, we investigate the application of artificial intelligence (AI) and in particular machine learning (ML) to the study of wireless propagation channels. It firstly provides a comprehensive overview of ML for channel characterization and ML-based antenna-channel optimization in this first part, and then it gives a state-of-the-art literature review of channel scenario identification and channel modeling in Part II. Fundamental results and key concepts of ML for communication networks are presented, and widely used ML methods for channel data processing, propagation channel estimation, and characterization are analyzed and compared. A discussion of challenges and future research directions for ML-enabled next generation networks of the topics covered in this part rounds off the paper.

eess.SP↗
arXiv2021

Artificial intelligence enabled radio propagation for communications-Part II: Scenario identification and channel modeling

This two-part paper investigates the application of artificial intelligence (AI) and in particular machine learning (ML) to the study of wireless propagation channels. In Part I, we introduced AI and ML as well as provided a comprehensive survey on ML enabled channel characterization and antenna-channel optimization, and in this part (Part II) we review state-of-the-art literature on scenario identification and channel modeling here. In particular, the key ideas of ML for scenario identification and channel modeling/prediction are presented, and the widely used ML methods for propagation scenario identification and channel modeling and prediction are analyzed and compared. Based on the state-of-art, the future challenges of AI/ML-based channel data processing techniques are given as well.

eess.SP↗
arXiv2021

A Geometry-Based Stochastic Model for Truck Communication Channels in Freeway Scenarios

Vehicle-to-vehicle (V2V) wireless communication systems are fundamental in many intelligent transportation applications, e.g., traffic load control, driverless vehicle, and collision avoidance. Hence, developing appropriate V2V communication systems and standardization require realistic V2V propagation channel models. However, most existing V2V channel modeling studies focus on car-to-car channels; only a few investigate truck-to-car (T2C) or truck-to-truck (T2T) channels. In this paper, a hybrid geometry-based stochastic model (GBSM) is proposed for T2X (T2C or T2T) channels in freeway environments. Next, we parameterize this GBSM from the extensive channel measurements. We extract the multipath components (MPCs) by using a joint maximum likelihood estimation (RiMAX) and then cluster the MPCs based on their evolution patterns.We classify the determined clusters as line-of-sight, multiple-bounce reflections from static interaction objects (IOs), multiple-bounce reflections from mobile IOs, multiple-bounce reflections, and diffuse scattering. Specifically, we model multiple-bounce reflections as double clusters following the COST 273/COST2100 method. This article presents the complete parameterization of the channel model. We validate this model by contrasting the root-mean-square delay spread and the angular spreads of departure/arrival derived from the channel model with the outcomes directly derived from the measurements.

eess.SP↗
arXiv2021

FASA: Feature Augmentation and Sampling Adaptation for Long-Tailed Instance Segmentation

Recent methods for long-tailed instance segmentation still struggle on rare object classes with few training data. We propose a simple yet effective method, Feature Augmentation and Sampling Adaptation (FASA), that addresses the data scarcity issue by augmenting the feature space especially for rare classes. Both the Feature Augmentation (FA) and feature sampling components are adaptive to the actual training status -- FA is informed by the feature mean and variance of observed real samples from past iterations, and we sample the generated virtual features in a loss-adapted manner to avoid over-fitting. FASA does not require any elaborate loss design, and removes the need for inter-class transfer learning that often involves large cost and manually-defined head/tail class groups. We show FASA is a fast, generic method that can be easily plugged into standard or long-tailed segmentation frameworks, with consistent performance gains and little added cost. FASA is also applicable to other tasks like long-tailed classification with state-of-the-art performance.

cs.CV↗
arXiv2021

An Attention Free Transformer

We introduce Attention Free Transformer (AFT), an efficient variant of Transformers that eliminates the need for dot product self attention. In an AFT layer, the key and value are first combined with a set of learned position biases, the result of which is multiplied with the query in an element-wise fashion. This new operation has a memory complexity linear w.r.t. both the context size and the dimension of features, making it compatible to both large input and model sizes. We also introduce AFT-local and AFT-conv, two model variants that take advantage of the idea of locality and spatial weight sharing while maintaining global connectivity. We conduct extensive experiments on two autoregressive modeling tasks (CIFAR10 and Enwik8) as well as an image recognition task (ImageNet-1K classification). We show that AFT demonstrates competitive performance on all the benchmarks, while providing excellent efficiency at the same time.

cs.LG↗
arXiv2021

Alignment Destabilizes Crystal Orders in Active Systems

We combine numerical and analytical methods to study two dimensional active crystals formed by permanently linked swimmers and with two distinct alignment interactions. The system admits a stationary phase with quasi long range translational order, as well as a moving phase with quasi-long range velocity order. The translational order in the moving phase is significantly influenced by alignment interaction. For Vicsek-like alignment, the translational order is short-ranged, whereas the bond-orientational order is quasi-long ranged, implying a moving hexatic phase. For elasticity-based alignment, the translational order is quasi-long ranged parallel to the motion and short-ranged in perpendicular direction, whereas the bond orientational order is long-ranged. We also generalize these results to higher dimensions.

cond-mat.soft↗
arXiv2021

MetricOpt: Learning to Optimize Black-Box Evaluation Metrics

We study the problem of directly optimizing arbitrary non-differentiable task evaluation metrics such as misclassification rate and recall. Our method, named MetricOpt, operates in a black-box setting where the computational details of the target metric are unknown. We achieve this by learning a differentiable value function, which maps compact task-specific model parameters to metric observations. The learned value function is easily pluggable into existing optimizers like SGD and Adam, and is effective for rapidly finetuning a pre-trained model. This leads to consistent improvements since the value function provides effective metric supervision during finetuning, and helps to correct the potential bias of loss-only supervision. MetricOpt achieves state-of-the-art performance on a variety of metrics for (image) classification, image retrieval and object detection. Solid benefits are found over competing methods, which often involve complex loss design or adaptation. MetricOpt also generalizes well to new tasks and model architectures.

cs.LG↗
arXiv2021

Radio Communication Scenarios in 5G-Railways

With the rapid development of railways, especially high-speed railways, there is an increasingly urgent demand for new wireless communication system for railways. Taking the mature 5G technology as an opportunity, 5G-railways (5G-R) have been widely regarded as a solution to meet the diversified demands of railway wireless communications. For the design, deployment and improvement of 5G-R networks, radio communication scenario classification plays an important role, affecting channel modeling and system performance evaluation. In this paper, a standardized radio communication scenario classification, including 18 scenarios, is proposed for 5G-R. This paper analyzes the differences of 5G-R scenarios compared with the traditional cellular networks and GSM-railways, according to 5G-R requirements and the unique physical environment and propagation characteristics. The proposed standardized scenario classification helps deepen the research of 5G-R and promote the development and application of the existing advanced technologies in railways.

cs.NI↗
arXiv2020

Information Swimmer: A Novel Mechanism of Self-propulsion

We study an information-based mechanism of self-propulsion in noisy environment. An information swimmer maintains directional motion by periodically measuring its velocity and accordingly adjusting its friction coefficient. Assuming that the measurement and adjustment are reversible and hence cause no energy dissipation, an information swimmer may move without external energy input. There is however no violation of the second law of thermodynamics, because the information entropy stored in the memory of swimmer increases monotonically. By optimizing its control parameters, the swimmer can achieve a steady velocity that is comparable to the root-mean-square velocity of an analogous Brownian particle. We also define a swimming efficiency in terms of information entropy production rate, and find that in equilibrium media with white noises, information swimmers are generally less efficient than Brownian particles driven by constant forces. For colored noises with long correlation times, the frequency of measurement can be greatly reduced without affecting the efficiency of information swimmers.

cond-mat.stat-mech↗