The Terahertz (0.1-10 THz) band holds enormous potential for supporting unprecedented data rates and millimeter-level accurate sensing thanks to its ultra-broad bandwidth. Terahertz integrated sensing and communication (ISAC) is viewed as a game-changing technology to realize connected intelligence in 6G and beyond systems. In this article, challenges from THz channel and transceiver perspectives, as well as difficulties of ISAC are elaborated. Motivated by these challenges, THz ISAC channels are studied in terms of channel types, measurement and models. Moreover, four key signal processing techniques to unleash the full potential of THz ISAC are investigated, namely, waveform design, receiver processing, narrowbeam management, and localization. Quantitative studies demonstrate the benefits and performance of the state-of-the-art signal processing methods. Finally, open problems and potential solutions are discussed.
Customer churn prediction is a valuable task in many industries. In telecommunications it presents great challenges, given the high dimensionality of the data, and how difficult it is to identify underlying frustration signatures, which may represent an important driver regarding future churn behaviour. Here, we propose a novel Bayesian hierarchical joint model that is able to characterise customer profiles based on how many events take place within different television watching journeys, and how long it takes between events. The model drastically reduces the dimensionality of the data from thousands of observations per customer to 11 customer-level parameter estimates and random effects. We test our methodology using data from 40 BT customers (20 active and 20 who eventually cancelled their subscription) whose TV watching behaviours were recorded from October to December 2019, totalling approximately half a million observations. Employing different machine learning techniques using the parameter estimates and random effects from the Bayesian hierarchical model as features yielded up to 92\% accuracy predicting churn, associated with 100\% true positive rates and false positive rates as low as 14\% on a validation set. Our proposed methodology represents an efficient way of reducing the dimensionality of the data, while at the same time maintaining high descriptive and predictive capabilities. We provide code to implement the Bayesian model at https://github.com/rafamoral/profiling_tv_watching_behaviour.
Zero-shot learning is a learning regime that recognizes unseen classes by generalizing the visual-semantic relationship learned from the seen classes. To obtain an effective ZSL model, one may resort to curating training samples from multiple sources, which may inevitably raise the privacy concerns about data sharing across different organizations. In this paper, we propose a novel Federated Zero-Shot Learning FedZSL framework, which learns a central model from the decentralized data residing on edge devices. To better generalize to previously unseen classes, FedZSL allows the training data on each device sampled from the non-overlapping classes, which are far from the i.i.d. that traditional federated learning commonly assumes. We identify two key challenges in our FedZSL protocol: 1) the trained models are prone to be biased to the locally observed classes, thus failing to generalize to the unseen classes and/or seen classes appeared on other devices; 2) as each category in the training data comes from a single source, the central model is highly vulnerable to model replacement (backdoor) attacks. To address these issues, we propose three local objectives for visual-semantic alignment and cross-device alignment through relation distillation, which leverages the normalized class-wise covariance to regularize the consistency of the prediction logits across devices. To defend against the backdoor attacks, a feature magnitude defending technique is proposed. As malicious samples are less correlated to the given semantic attributes, the visual features of low magnitude will be discarded to stabilize model updates. The effectiveness and robustness of FedZSL are demonstrated by extensive experiments conducted on three zero-shot benchmark datasets.
Modeling users' dynamic preferences from historical behaviors lies at the core of modern recommender systems. Due to the diverse nature of user interests, recent advances propose the multi-interest networks to encode historical behaviors into multiple interest vectors. In real scenarios, the corresponding items of captured interests are usually retrieved together to get exposure and collected into training data, which produces dependencies among interests. Unfortunately, multi-interest networks may incorrectly concentrate on subtle dependencies among captured interests. Misled by these dependencies, the spurious correlations between irrelevant interests and targets are captured, resulting in the instability of prediction results when training and test distributions do not match. In this paper, we introduce the widely used Hilbert-Schmidt Independence Criterion (HSIC) to measure the degree of independence among captured interests and empirically show that the continuous increase of HSIC may harm model performance. Based on this, we propose a novel multi-interest network, named DEep Stable Multi-Interest Learning (DESMIL), which tries to eliminate the influence of subtle dependencies among captured interests via learning weights for training samples and make model concentrate more on underlying true causation. We conduct extensive experiments on public recommendation datasets, a large-scale industrial dataset and the synthetic datasets which simulate the out-of-distribution data. Experimental results demonstrate that our proposed DESMIL outperforms state-of-the-art models by a significant margin. Besides, we also conduct comprehensive model analysis to reveal the reason why DESMIL works to a certain extent.
Generalized Zero-Shot Learning (GZSL) aims to recognize images from both the seen and unseen classes by transferring semantic knowledge from seen to unseen classes. It is a promising solution to take the advantage of generative models to hallucinate realistic unseen samples based on the knowledge learned from the seen classes. However, due to the generation shifts, the synthesized samples by most existing methods may drift from the real distribution of the unseen data. To address this issue, we propose a novel flow-based generative framework that consists of multiple conditional affine coupling layers for learning unseen data generation. Specifically, we discover and address three potential problems that trigger the generation shifts, i.e., semantic inconsistency, variance collapse, and structure disorder. First, to enhance the reflection of the semantic information in the generated samples, we explicitly embed the semantic information into the transformation in each conditional affine coupling layer. Second, to recover the intrinsic variance of the real unseen features, we introduce a boundary sample mining strategy with entropy maximization to discover more difficult visual variants of semantic prototypes and hereby adjust the decision boundary of the classifiers. Third, a relative positioning strategy is proposed to revise the attribute embeddings, guiding them to fully preserve the inter-class geometric structure and further avoid structure disorder in the semantic space. Extensive experimental results on four GZSL benchmark datasets demonstrate that GSMFlow achieves the state-of-the-art performance on GZSL.
The combination of cell-free massive multiple-input multiple-output (CF-mMIMO) and reconfigurable intelligent surface (RIS) is envisioned as a promising paradigm to improve network capacity and enhance coverage capability. However, to reap full benefits of RIS-aided CF-mMIMO, the main challenge is to efficiently design cooperative beamforming (CBF) at base stations (BSs), RISs, and users. Firstly, we investigate the fractional programing to convert the weighted sum-rate (WSR) maximization problem into a tractable optimization problem. Then, the alternating optimization framework is employed to decompose the transformed problem into a sequence of subproblems, i.e., hybrid BF (HBF) at BSs, passive BF at RISs, and combining at users. In particular, the alternating direction method of multipliers algorithm is utilized to solve the HBF subproblem at BSs. Concretely, the analog BF design with unit-modulus constraints is solved by the manifold optimization (MO) while we obtain a closed-form solution to the digital BF design that is essentially a convex least-square problem. Additionally, the passive BF at RISs and the analog combining at users are designed by primal-dual subgradient and MO methods. Moreover, considering heavy communication costs in conventional CF-mMIMO systems, we propose a partially-connected CF-mMIMO (P-CF-mMIMO) framework to decrease the number of connections among BSs and users. To better compromise WSR performance and network costs, we formulate the BS selection problem in the P-CF-mMIMO system as a binary integer quadratic programming (BIQP) problem, and develop a relaxed linear approximation algorithm to handle this BIQP problem. Finally, numerical results demonstrate superiorities of our proposed algorithms over baseline counterparts.
Reconfigurable intelligent surface (RIS) and hybrid beamforming have been envisioned as promising alternatives to alleviate blockage vulnerability and enhance coverage capability for terahertz (THz) multi-user multiple-input multiple-output systems that suffer from severe propagation attenuation and poor diffraction. Considering that the joint beamforming with large-scale array elements at transceivers and RIS is extremely complicated, the codebook based beamforming can be employed in a computationally efficient manner. However, the codeword selection for analog beamforming is an intractable combinatorial optimization (CO) problem. To this end, an iterative alternating search (IAS) algorithm is developed to achieve a near-optimal sum-rate performance with low computational complexity in contrast with the optimal exhaustive search algorithm. According to the THz channel dataset generated by the IAS algorithm, a multi-task learning based analog beam selection (MTL-ABS) framework is developed to further decrease the computation overhead. Specifically, we take the CO problem as a multi-task classification problem and implement multiple beam selection tasks at transceivers and RIS simultaneously. Remarkably, residual network and self-attention mechanism are used to combat the network degradation and mine intrinsic THz channel features. Finally, blockwise convergence analysis and numerical results demonstrate the effectiveness of the MTL-ABS framework over search based counterparts.
Terahertz (THz) communications have been envisioned as a promising enabler to provide ultra-high data transmission for sixth generation (6G) wireless networks. To tackle the blockage vulnerability brought by severe attenuation and poor diffraction of THz waves, a nanoscale reconfigurable intelligent surface (NRIS) is developed to smartly manipulate the propagation directions of incident THz waves. In this paper, the electric properties of the graphene are investigated by revealing the relationship between conductivity and applied voltages, and then an efficient hardware structure of electrically-controlled NRIS is designed based on Fabry-Perot resonance model. Particularly, the phase response of NRIS can be programmed up to 306.82 degrees. To analyze the hardware performance, we jointly design the passive and active beamforming for NRIS aided THz communication system. Particularly, an adaptive gradient descent (A-GD) algorithm is developed to optimize the phase shift matrix of NRIS by dynamically updating the step size during the iterative process. Finally, numerical results demonstrate the effectiveness of our designed hardware architecture as well as the developed algorithm.
In industry, feature selection is a standard but necessary step to search for an optimal set of informative feature fields for efficient and effective training of deep Click-Through Rate (CTR) models. Most previous works measure the importance of feature fields by using their corresponding continuous weights from the model, then remove the feature fields with small weight values. However, removing many features that correspond to small but not exact zero weights will inevitably hurt model performance and not be friendly to hot-start model training. There is also no theoretical guarantee that the magnitude of weights can represent the importance, thus possibly leading to sub-optimal results if using these methods. To tackle this problem, we propose a novel Learnable Polarizing Feature Selection (LPFS) method using a smoothed-$\ell^0$ function in literature. Furthermore, we extend LPFS to LPFS++ by our newly designed smoothed-$\ell^0$-liked function to select a more informative subset of features. LPFS and LPFS++ can be used as gates inserted at the input of the deep network to control the active and inactive state of each feature. When training is finished, some gates are exact zero, while others are around one, which is particularly favored by the practical hot-start training in the industry, due to no damage to the model performance before and after removing the features corresponding to exact-zero gates. Experiments show that our methods outperform others by a clear margin, and have achieved great A/B test results in KuaiShou Technology.