Abstract:Do neural network models of vision learn brain-aligned representations because they share architectural constraints and task objectives with biological vision or because they learn universal features of natural image processing? We characterized the universality of hundreds of thousands of representational dimensions from visual neural networks with varied construction. We found that networks with varied architectures and task objectives learn to represent natural images using a shared set of latent dimensions, despite appearing highly distinct at a surface level. Next, by comparing these networks with human brain representations measured with fMRI, we found that the most brain-aligned representations in neural networks are those that are universal and independent of a network's specific characteristics. Remarkably, each network can be reduced to fewer than ten of its most universal dimensions with little impact on its representational similarity to the human brain. These results suggest that the underlying similarities between artificial and biological vision are primarily governed by a core set of universal image representations that are convergently learned by diverse systems.
Abstract:The integration of Large Language Models (LLMs) with Knowledge Representation Learning (KRL) signifies a pivotal advancement in the field of artificial intelligence, enhancing the ability to capture and utilize complex knowledge structures. This synergy leverages the advanced linguistic and contextual understanding capabilities of LLMs to improve the accuracy, adaptability, and efficacy of KRL, thereby expanding its applications and potential. Despite the increasing volume of research focused on embedding LLMs within the domain of knowledge representation, a thorough review that examines the fundamental components and processes of these enhanced models is conspicuously absent. Our survey addresses this by categorizing these models based on three distinct Transformer architectures, and by analyzing experimental data from various KRL downstream tasks to evaluate the strengths and weaknesses of each approach. Finally, we identify and explore potential future research directions in this emerging yet underexplored domain, proposing pathways for continued progress.
Abstract:Recently, studies have shown the potential of integrating field-type iterative methods with deep learning (DL) techniques in solving inverse scattering problems (ISPs). In this article, we propose a novel Variational Born Iterative Network, namely, VBIM-Net, to solve the full-wave ISPs with significantly improved flexibility and inversion quality. The proposed VBIM-Net emulates the alternating updates of the total electric field and the contrast in the variational Born iterative method (VBIM) by multiple layers of subnetworks. We embed the calculation of the contrast variation into each of the subnetworks, converting the scattered field residual into an approximate contrast variation and then enhancing it by a U-Net, thus avoiding the requirement of matched measurement dimension and grid resolution as in existing approaches. The total field and contrast of each layer's output is supervised in the loss function of VBIM-Net, which guarantees the physical interpretability of variables of the subnetworks. In addition, we design a training scheme with extra noise to enhance the model's stability. Extensive numerical results on synthetic and experimental data both verify the inversion quality, generalization ability, and robustness of the proposed VBIM-Net. This work may provide some new inspiration for the design of efficient field-type DL schemes.
Abstract:How to reduce the pilot overhead required for channel estimation? How to deal with the channel dynamic changes and error propagation in channel prediction? To jointly address these two critical issues in next-generation transceiver design, in this paper, we propose a novel framework named channel deduction for high-dimensional channel acquisition in multiple-input multiple-output (MIMO)-orthogonal frequency division multiplexing (OFDM) systems. Specifically, it makes use of the outdated channel information of past time slots, performs coarse estimation for the current channel with a relatively small number of pilots, and then fuses these two information to obtain a complete representation of the present channel. The rationale is to align the current channel representation to both the latent channel features within the past samples and the coarse estimate of current channel at the pilots, which, in a sense, behaves as a complementary combination of estimation and prediction and thus reduces the overall overhead. To fully exploit the highly nonlinear correlations in time, space, and frequency domains, we resort to learning-based implementation approaches. By using the highly efficient complex-domain multilayer perceptron (MLP)-mixer for crossing space-frequency domain representation and the recurrence-based or attention-based mechanisms for the past-present interaction, we respectively design two different channel deduction neural networks (CDNets). We provide a general procedure of data collection, training, and deployment to standardize the application of CDNets. Comprehensive experimental evaluations in accuracy, robustness, and efficiency demonstrate the superiority of the proposed approach, which reduces the pilot overhead by up to 88.9% compared to state-of-the-art estimation approaches and enables continuous operating even under unknown user movement and error propagation.
Abstract:In multiple-input multiple-output (MIMO) orthogonal frequency division multiplexing (OFDM) systems, representing the whole channel only based on partial subchannels will significantly reduce the channel acquisition overhead. For such a channel mapping task, inspired by the intrinsic coupling across the space and frequency domains, this letter proposes to use interleaved learning with partial antenna and subcarrier characteristics to represent the whole MIMO-OFDM channel. Specifically, we design a complex-domain multilayer perceptron (MLP)-Mixer (CMixer), which utilizes two kinds of complex-domain MLP modules to learn the space and frequency characteristics respectively and then interleaves them to couple the learned properties. The complex-domain computation facilitates the learning on the complex-valued channel data, while the interleaving tightens the coupling of space and frequency domains. These two designs jointly reduce the learning burden, making the physics-inspired CMixer more effective on channel representation learning than existing data-driven approaches. Simulation shows that the proposed scheme brings 4.6~10dB gains in mapping accuracy compared to existing schemes under different settings. Besides, ablation studies show the necessity of complex-domain computation as well as the extent to which the interleaved learning matches the channel properties.
Abstract:Knowledge graphs generally suffer from incompleteness, which can be alleviated by completing the missing information. Deep knowledge convolutional embedding models based on neural networks are currently popular methods for knowledge graph completion. However, most existing methods use external convolution kernels and traditional plain convolution processes, which limits the feature interaction capability of the model. In this paper, we propose a novel dynamic convolutional embedding model ConvD for knowledge graph completion, which directly reshapes the relation embeddings into multiple internal convolution kernels to improve the external convolution kernels of the traditional convolutional embedding model. The internal convolution kernels can effectively augment the feature interaction between the relation embeddings and entity embeddings, thus enhancing the model embedding performance. Moreover, we design a priori knowledge-optimized attention mechanism, which can assign different contribution weight coefficients to multiple relation convolution kernels for dynamic convolution to improve the expressiveness of the model further. Extensive experiments on various datasets show that our proposed model consistently outperforms the state-of-the-art baseline methods, with average improvements ranging from 11.30\% to 16.92\% across all model evaluation metrics. Ablation experiments verify the effectiveness of each component module of the ConvD model.
Abstract:Recently, big artificial intelligence (AI) models represented by chatGPT have brought an incredible revolution. With the pre-trained big AI model (BAIM) in certain fields, numerous downstream tasks can be accomplished with only few-shot or even zero-shot learning and exhibit state-of-the-art performances. As widely envisioned, the big AI models are to rapidly penetrate into major intelligent services and applications, and are able to run at low unit cost and high flexibility. In 6G wireless networks, to fully enable intelligent communication, sensing and computing, apart from providing other intelligent wireless services and applications, it is of vital importance to design and deploy certain wireless BAIMs (wBAIMs). However, there still lacks investigation on architecture design and system evaluation for wBAIM. In this paper, we provide a comprehensive discussion as well as some in-depth prospects on the demand, design and deployment aspects of the wBAIM. We opine that wBAIM will be a recipe of the 6G wireless networks to build high-efficient, sustainable, versatile, and extensible wireless intelligence for numerous promising visions. Then, we present the core characteristics and principles to guide the design of wBAIMs, and discuss the key aspects of developing wBAIMs through identifying the differences between the existing BAIMs and the emerging wBAIMs. Finally, related research directions and potential solutions are outlined.
Abstract:In this paper, we propose an innovative learning-based channel prediction scheme so as to achieve higher prediction accuracy and reduce the requirements of huge amount and strict sequential format of channel data. Inspired by the idea of the neural ordinary differential equation (Neural ODE), we first prove that the channel prediction problem can be modeled as an ODE problem with a known initial value through analyzing the physical process of electromagnetic wave propagation within a varying space. Then, we design a novel physics-inspired spatial channel gradient network (SCGNet), which represents the derivative process of channel varying as a special neural network and can obtain the gradients at any relative displacement needed for the ODE solving. With the SCGNet, the static channel at any location served by the base station is accurately inferred through consecutive propagation and integration. Finally, we design an efficient recurrent positioning algorithm based on some prior knowledge of user mobility to obtain the velocity vector, and propose an approximate Doppler compensation method to make up the instantaneous angular-delay domain channel. Only discrete historical channel data is needed for the training, whereas only a few fresh channel measurements is needed for the prediction, which ensures the scheme's practicability.
Abstract:Obtaining accurate channel state information (CSI) is crucial and challenging for multiple-input multiple-output (MIMO) wireless communication systems. Conventional channel estimation method cannot guarantee the accuracy of mobile CSI while requires high signaling overhead. Through exploring the intrinsic correlation among a set of historical CSI instances randomly obtained in a certain communication environment, channel prediction can significantly increase CSI accuracy and save signaling overhead. In this paper, we propose a novel channel prediction method based on ordinary differential equation (ODE)-recurrent neural network (RNN) for accurate and flexible mobile MIMO channel prediction. Differing from existing works using sequential network structures for exploring the numerical correlation between observed data, our proposed method tries to represent the implicit physics process of path responses changing by specially designed continuous learning network with ODE structure. Due to the targeted design of learning network, our proposed method fits the mathematics feature of CSI data better and enjoy higher network interpretability. Experimental results show that the proposed learning approach outperforms existing methods, especially for long time interval of the CSI sequence and large channel measurement error.