Xuanyu Lei, Chenliang Li, Yuning Wu, Kaiming Liu, Weizhou Shen, Peng Li, Ming Yan, Fei Huang, Ya-Qin Zhang, Yang Liu. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

The Triplet Network has made significant advancements and garnered considerable attention from both academia and industry. With the rapid development of wireless technologies, the vast amounts of data generated at the network edge present an unprecedented opportunity for the application of Triplet Network. However, due to constraints such as limited bandwidth, privacy concerns, and legal challenges, it is impractical to transmit all data to the cloud or central servers for analysis, training, and processing. As a result, deploying and training Triplet Network at the edge has emerged as a promising alternative solution. Federated learning (FL), an emerging distributed learning paradigm, offers potential solutions to this challenge by enabling model training while preserving data privacy. However, due to the reliance of traditional Triplet Network on fixed input formats, the flexibility of the model is limited. In this paper, we enhance the model’s flexibility by altering the input format of the Triplet Network and introducing an Adaptive Weighted Triplet Loss. Then, to further improve the feature extraction capability of the Triplet Network, we incorporate the CBAM attention module. Theoretical analysis and experimental results demonstrate that the improved Triplet Network significantly enhances the accuracy of image classification tasks and accelerates model convergence. Experimental results on public datasets show the effectiveness of the proposed framework.

The rapid progress of foundation models has led to the prosperity of autonomous agents, which leverage the universal capabilities of foundation models to conduct reasoning, decision-making, and environmental interaction. However, the efficacy of agents remains limited when operating in intricate, realistic environments. In this work, we introduce the principles of Unified Alignment for Agents (UA$^2$), which advocate for the simultaneous alignment of agents with human intentions, environmental dynamics, and self-constraints such as the limitation of monetary budgets. From the perspective of UA$^2$, we review the current agent research and highlight the neglected factors in existing agent benchmarks and method candidates. We also conduct proof-of-concept studies by introducing realistic features to WebShop, including user profiles demonstrating intentions, personalized reranking reflecting complex environmental dynamics, and runtime cost statistics as self-constraints. We then follow the principles of UA$^2$ to propose an initial design of our agent and benchmark its performance with several candidate baselines in the retrofitted WebShop. The extensive experimental results further prove the importance of the principles of UA$^2$. Our research sheds light on the next steps of autonomous agent research with improved general problem-solving abilities.

With the massive emergence of multi-modal data, cross-modal retrieval (CMR) has become one of the hot topics. Thanks to fast retrieval and efficient storage, cross-modal hashing (CMH) provides a feasible solution for large-scale multi-modal data. Previous CMH methods always directly learn common hash codes to fuse different modalities. Although they have obtained some success, there are still some limitations: 1) These approaches often prioritize reducing the heterogeneity in multi-modal data by learning consensus hash codes, yet they could sacrifice modality-specific information. 2) They frequently utilize pairwise similarities to guide hashing learning and neglect class distribution correlations. To overcome these two issues, we propose a novel Distribution Consistency Guided Hashing (DCGH) framework. Specifically, we first learn the modality-specific representation to extract the private discriminative information. Further, we learn consensus hash codes from the private representation by consensus hashing learning, thereby merging the specifics with consistency. Finally, we propose distribution consistency learning to guide hash codes following a similar class distribution principle between multi-modal data, thereby exploring more consistent information. Lots of experimental results on four benchmark datasets demonstrate the effectiveness of our DCGH on both fully paired and partially paired CMR tasks. The code can be available at: https://github.com/sunyuan-cs/2024-MM-DCGH.

Orthogonal Time Frequency Space (OTFS) modulation proposed in recent years has received much attention due to its ability to maintain good performance in environments with high frequency dispersion. However, the OTFS system has an inherent high peak-to-average power ratio (PAPR) problem, which impairs the efficiency of power amplifiers. In this paper, we propose a novel PAPR reduction method for OTFS systems, which first introduces the grid reservation (GR) scheme in the delay-Doppler (DD) domain. This proposed method reserves a small portion of the grids for carrying the preset clipping signals (PCS), thus reducing the signal PAPR while introducing none distortions to original signals. Simulation results indicate that the PAPR of OTFS signals can be reduced significantly at the cost of little transmission resources, without degradation of the bit error rate (BER) performance.

In emergency communication scenarios, task nodes often need to perform rescue missions in areas lacking basic network coverage. To ensure reliable communication, emergency communication relay networks are constructed, necessitating relay nodes to establish multiple relay links and dynamically adjust the network topology and the number of nodes. Traditional relay deployment schemes, which rely on heuristic algorithms, often lack flexibility and may become trapped in local optima. To address these issues, we propose an Intelligent Relay Nodes Deployment Method (INDM). This method combines Twin Delayed Deep Deterministic Policy Gradient (TD3) and Double Deep Q-Network (DQN) algorithms to optimize the number and placement of nodes in the relay link. Additionally, an adaptive segmented reward function is designed to guide agent iterations, while noise exploration mechanisms and an adaptive experience replay algorithm enable agents to fully explore and identify optimal strategies. Simulation results demonstrate that our algorithm effectively reduces relay node energy consumption and the total number of nodes, all while ensuring high communication quality for task nodes.

In this study, the authors have proposed a low complexity companding function for peak‐to‐average‐power ratio (PAPR) reduction in orthogonal frequency division multiplexing systems. This function transforms the signals into a probability density function with a cumulative distribution function an anti‐trigonometric function. By carefully choosing the transform parameters, the bit error rate (BER) performance can be improved significantly during the reduction of PAPR. In addition, the proposed companding function offers design flexibility. Simulation results show that for the same PAPR performance, the proposed companding scheme can offer better BER performance. For example, upon assigning BER = 10 −4 in an additive white Gaussian noise channel without a solid‐state power amplifier, the minimum required bit energy to noise power spectral ratio ( E b / N o ) for the proposed companding scheme with c = 0 and A = 1.414 is 11 dB. It is 5, 4.2, and 3 dB lower than exponential companding function, root‐based μ ‐law companding function, and amplitude limiting companding schemes, respectively. Also, their PAPR performance is the same.

Energy detection is a widely used method for spectrum sensing in cognitive radios because of its simplicity and accuracy. However, it is severely affected by noise uncertainty. To solve this problem, this study raises a blind multiband spectrum sensing method with smart antennas which is robust to noise uncertainty. The proposed method performs spectrum sensing simultaneously over the total frequency channels rather than a single channel each time. Owing to the fact that the noise eigenvectors of the sample covariance matrix are orthogonal to the direction matrix whereas the signal eigenvectors are not, the proposal utilises the Gerschgorin radii of the unitary transformed covariance matrix to distinguish the noise from the signal. Unlike the conventional sensing methods, the authors' approach does not need any prior knowledge of the noise power or the primary user signals, which makes it suitable for blind spectrum sensing. This study presents simulations in various conditions to validate the performance of the proposed scheme and shows that their proposal outperforms the other existing methods.

This study mainly focuses on the K user multiple input multiple output (MIMO) Gaussian interference channel with M antennas at each transmitter and N antennas at each receiver, in which min( M , N )/max( M , N ) ≥ ( K − 1)/[ K ( K − 2)] and K > 3. The channel coefficients are time varying/frequency selective and drawn from a continuous distribution. Based on ergodic interference alignment (IA), an achievable scheme is proposed to achieve the KMN /( M + N ) total degrees of freedom (DoF). Benefited from ergodic IA, the scheme can reach the goal with simply linear beamforming and finite symbols. Nevertheless, the limitation of ergodic IA is the long delay waiting for the desired channels. Hence, the actual delay is derived. Then, by multiplying factors on channel coefficients and adjusting orders of symbols, the authors optimise the delay. The formula shows that the new delay can be effectively reduced. In addition, the simplicity of coding scheme allows us to derive the closed form of achievable rate expression at any signal‐to‐noise ratio. And the optimisation of latency does not affect the data rate at all. The simulation shows that the achievable rate is close to the rate upper bound.

Prevalence of real-time service in Internet and mobile ad hoc network (MANET) is bound to stimulate the technical progress in the interconnection of MANET and Internet. The interconnection is achieved by using Internet gateways, which act as bridges between MANET and Internet. When a node in MANET wants to obtain a globally routable address for Internet connection, it has to first find route to a gateway. In some network conditions, performance of Internet connectivity will be degraded by obvious unidirectional links and broadcast storm due to the blind rebroadcast of gateway discovery packets. In this paper, an effective hybrid gateway discovery algorithm for MANET is proposed to remove unidirectional links from global route computation simultaneously and enhance Internet connectivity. The routing protocol called ad hoc on-demand distance vector (AODV) has been modified to set up support hybrid network. Results using NS2 simulation show this hybrid Internet connectivity approach can provide better performance than others due to avoiding unidirectional links with little overhead.

Content-based image retrieval (CBIR) on the Web is an active research topic in recent years. However, a common problem for the content-based search is the difficulty for getting a query image, which dramatically limits the popularity of their application. The ubiquity of camera phones has opened up a new avenue for image-based mobile search and it also gives new challenges. In this paper, a domain-specific CBIR system on the Web with mobile camera-phone is proposed, which further extends our previous work on VAST system and focuses on the specific garment images. It makes full use of the camera-phone's ability that captures and views images anywhere. The key problems related to the wireless network and media characteristics, system usability and users expectations are depicted. The system architecture and flowchart are described also. Based on these, a prototype is implemented.

In this paper, we first give an overview of security architecture of mobile WiMAX network. Then, we investigate man-in-the-middle attacks and Denial of Service (DoS) attacks toward 802.16e-based Mobile WiMAX network. We find the initial network procedure is not effectively secured that makes Man-in-the-middle and Dos attacks possible. In addition, we find the resource saving and handover procedure is not secured enough to resist DoS attacks. Focusing on these two kinds of attacks, we propose Secure Initial Network Entry Protocol (SINEP) based on Diffie-Hellman (DH) key exchange protocol to enhance the security level during network initial. We modify DH key exchange protocol to fit it into mobile WiMAX network as well as to eliminate existing weakness in original DH key exchange protocol.

A crucial aspect for adaptive bit and power loading algorithms is to determine all usable subchannels. In this paper, a simplified bandwidth optimization method is proposed to search for all usable subchannels for loading algorithms which aim at the minimization of power consumption with constraint of constant data rate and target bit error rate (BER). Based on this method, an efficient loading algorithm is developed. Simulation results show that the proposed algorithm can achieve the same bit allocation results to that of the well-known Hughes-Hartogs and Fischer et al. algorithm. This algorithm is appropriate for OFDM systems in time-varying channels due to its low complexity.

This paper presents a new efficient bit and power loading algorithm for orthogonal frequency division multiplexing (OFDM) systems, which aims to minimize the transmit power while maintaining a constant data rate and target bit error rate (BER). This algorithm is based on greedy allocation approach. After grouping, sorting and clustering, the greedy allocation procedure is performed over clusters other than over all subchannels. Compared with conventional optimal loading algorithms, the proposed algorithm can reduce the algorithmic complexity significantly, and the performance degradation may be negligible if algorithmic parameters are selected appropriately. Simulation results indicate advantages of the proposed algorithm.

In this paper a simplified differential amplitude demodulation scheme is proposed for 64-DAPSK (differential amplitude and phase shift keying), which needs less quantity of decision thresholds compared with the conventional schemes. The arithmetic mean threshold (AlMT) scheme is proposed and analyzed with the simplified system in AWGN and flat Rayleigh fading channels. By these proposed schemes, system complexity is reduced considerably, but no performance degradation occurs. In the case of channel coding, the concatenated code consisting of an inner convolutional code and an outer Reed-Solomon code can be applied to 64-DAPSK. Compared with coherent 64-QAM system which employs pilot symbols for channel estimation, 64- DAPSK can achieve the performance close to or even better than that of the former by utilizing the additional redundancy for outer coding, provided that the channel SNR is high enough.