Token-level Direct Preference Optimization

Fine-tuning pre-trained Large Language Models (LLMs) is essential to align them with human values and intentions. This process often utilizes methods like pairwise comparisons and KL divergence against a reference LLM, focusing on the evaluation of full answers generated by the models. However, the generation of these responses occurs in a token level, following a sequential, auto-regressive fashion. In this paper, we introduce Token-level Direct Preference Optimization (TDPO), a novel approach to align LLMs with human preferences by optimizing policy at the token level. Unlike previous methods, which face challenges in divergence efficiency, TDPO incorporates forward KL divergence constraints for each token, improving alignment and diversity. Utilizing the Bradley-Terry model for a token-based reward system, TDPO enhances the regulation of KL divergence, while preserving simplicity without the need for explicit reward modeling. Experimental results across various text tasks demonstrate TDPO's superior performance in balancing alignment with generation diversity. Notably, fine-tuning with TDPO strikes a better balance than DPO in the controlled sentiment generation and single-turn dialogue datasets, and significantly improves the quality of generated responses compared to both DPO and PPO-based RLHF methods. Our code is open-sourced at https://github.com/Vance0124/Token-level-Direct-Preference-Optimization.

Adaptive Fair Representation Learning for Personalized Fairness in Recommendations via Information Alignment

Personalized fairness in recommendations has been attracting increasing attention from researchers. The existing works often treat a fairness requirement, represented as a collection of sensitive attributes, as a hyper-parameter, and pursue extreme fairness by completely removing information of sensitive attributes from the learned fair embedding, which suffer from two challenges: huge training cost incurred by the explosion of attribute combinations, and the suboptimal trade-off between fairness and accuracy. In this paper, we propose a novel Adaptive Fair Representation Learning (AFRL) model, which achieves a real personalized fairness due to its advantage of training only one model to adaptively serve different fairness requirements during inference phase. Particularly, AFRL treats fairness requirements as inputs and can learn an attribute-specific embedding for each attribute from the unfair user embedding, which endows AFRL with the adaptability during inference phase to determine the non-sensitive attributes under the guidance of the user's unique fairness requirement. To achieve a better trade-off between fairness and accuracy in recommendations, AFRL conducts a novel Information Alignment to exactly preserve discriminative information of non-sensitive attributes and incorporate a debiased collaborative embedding into the fair embedding to capture attribute-independent collaborative signals, without loss of fairness. Finally, the extensive experiments conducted on real datasets together with the sound theoretical analysis demonstrate the superiority of AFRL.

Multi-objective Deep Reinforcement Learning for Mobile Edge Computing

Mobile edge computing (MEC) is essential for next-generation mobile network applications that prioritize various performance metrics, including delays and energy consumption. However, conventional single-objective scheduling solutions cannot be directly applied to practical systems in which the preferences of these applications (i.e., the weights of different objectives) are often unknown or challenging to specify in advance. In this study, we address this issue by formulating a multi-objective offloading problem for MEC with multiple edges to minimize expected long-term energy consumption and transmission delay while considering unknown preferences as parameters. To address the challenge of unknown preferences, we design a multi-objective (deep) reinforcement learning (MORL)-based resource scheduling scheme with proximal policy optimization (PPO). In addition, we introduce a well-designed state encoding method for constructing features for multiple edges in MEC systems, a sophisticated reward function for accurately computing the utilities of delay and energy consumption. Simulation results demonstrate that our proposed MORL scheme enhances the hypervolume of the Pareto front by up to 233.1% compared to benchmarks. Our full framework is available at https://github.com/gracefulning/mec_morl_multipolicy.

Minimizing Age of Information for Mobile Edge Computing Systems: A Nested Index Approach

Exploiting the computational heterogeneity of mobile devices and edge nodes, mobile edge computation (MEC) provides an efficient approach to achieving real-time applications that are sensitive to information freshness, by offloading tasks from mobile devices to edge nodes. We use the metric Age-of-Information (AoI) to evaluate information freshness. An efficient solution to minimize the AoI for the MEC system with multiple users is non-trivial to obtain due to the random computing time. In this paper, we consider multiple users offloading tasks to heterogeneous edge servers in a MEC system. We first reformulate the problem as a Restless Multi-Arm-Bandit (RMAB) problem and establish a hierarchical Markov Decision Process (MDP) to characterize the updating of AoI for the MEC system. Based on the hierarchical MDP, we propose a nested index framework and design a nested index policy with provably asymptotic optimality. Finally, the closed form of the nested index is obtained, which enables the performance tradeoffs between computation complexity and accuracy. Our algorithm leads to an optimality gap reduction of up to 40%, compared to benchmarks. Our algorithm asymptotically approximates the lower bound as the system scalar gets large enough.

Contrastive Collaborative Filtering for Cold-Start Item Recommendation

The cold-start problem is a long-standing challenge in recommender systems. As a promising solution, content-based generative models usually project a cold-start item's content onto a warm-start item embedding to capture collaborative signals from item content so that collaborative filtering can be applied. However, since the training of the cold-start recommendation models is conducted on warm datasets, the existent methods face the issue that the collaborative embeddings of items will be blurred, which significantly degenerates the performance of cold-start item recommendation. To address this issue, we propose a novel model called Contrastive Collaborative Filtering for Cold-start item Recommendation (CCFCRec), which capitalizes on the co-occurrence collaborative signals in warm training data to alleviate the issue of blurry collaborative embeddings for cold-start item recommendation. In particular, we devise a contrastive collaborative filtering (CF) framework, consisting of a content CF module and a co-occurrence CF module to generate the content-based collaborative embedding and the co-occurrence collaborative embedding for a training item, respectively. During the joint training of the two CF modules, we apply a contrastive learning between the two collaborative embeddings, by which the knowledge about the co-occurrence signals can be indirectly transferred to the content CF module, so that the blurry collaborative embeddings can be rectified implicitly by the memorized co-occurrence collaborative signals during the applying phase. Together with the sound theoretical analysis, the extensive experiments conducted on real datasets demonstrate the superiority of the proposed model. The codes and datasets are available on https://github.com/zzhin/CCFCRec.

Provable Unrestricted Adversarial Training without Compromise with Generalizability

Adversarial training (AT) is widely considered as the most promising strategy to defend against adversarial attacks and has drawn increasing interest from researchers. However, the existing AT methods still suffer from two challenges. First, they are unable to handle unrestricted adversarial examples (UAEs), which are built from scratch, as opposed to restricted adversarial examples (RAEs), which are created by adding perturbations bound by an $l_p$ norm to observed examples. Second, the existing AT methods often achieve adversarial robustness at the expense of standard generalizability (i.e., the accuracy on natural examples) because they make a tradeoff between them. To overcome these challenges, we propose a unique viewpoint that understands UAEs as imperceptibly perturbed unobserved examples. Also, we find that the tradeoff results from the separation of the distributions of adversarial examples and natural examples. Based on these ideas, we propose a novel AT approach called Provable Unrestricted Adversarial Training (PUAT), which can provide a target classifier with comprehensive adversarial robustness against both UAE and RAE, and simultaneously improve its standard generalizability. Particularly, PUAT utilizes partially labeled data to achieve effective UAE generation by accurately capturing the natural data distribution through a novel augmented triple-GAN. At the same time, PUAT extends the traditional AT by introducing the supervised loss of the target classifier into the adversarial loss and achieves the alignment between the UAE distribution, the natural data distribution, and the distribution learned by the classifier, with the collaboration of the augmented triple-GAN. Finally, the solid theoretical analysis and extensive experiments conducted on widely-used benchmarks demonstrate the superiority of PUAT.

Boundary Guided Semantic Learning for Real-time COVID-19 Lung Infection Segmentation System

The coronavirus disease 2019 (COVID-19) continues to have a negative impact on healthcare systems around the world, though the vaccines have been developed and national vaccination coverage rate is steadily increasing. At the current stage, automatically segmenting the lung infection area from CT images is essential for the diagnosis and treatment of COVID-19. Thanks to the development of deep learning technology, some deep learning solutions for lung infection segmentation have been proposed. However, due to the scattered distribution, complex background interference and blurred boundaries, the accuracy and completeness of the existing models are still unsatisfactory. To this end, we propose a boundary guided semantic learning network (BSNet) in this paper. On the one hand, the dual-branch semantic enhancement module that combines the top-level semantic preservation and progressive semantic integration is designed to model the complementary relationship between different high-level features, thereby promoting the generation of more complete segmentation results. On the other hand, the mirror-symmetric boundary guidance module is proposed to accurately detect the boundaries of the lesion regions in a mirror-symmetric way. Experiments on the publicly available dataset demonstrate that our BSNet outperforms the existing state-of-the-art competitors and achieves a real-time inference speed of 44 FPS.

Time Lag Aware Sequential Recommendation

Although a variety of methods have been proposed for sequential recommendation, it is still far from being well solved partly due to two challenges. First, the existing methods often lack the simultaneous consideration of the global stability and local fluctuation of user preference, which might degrade the learning of a user's current preference. Second, the existing methods often use a scalar based weighting schema to fuse the long-term and short-term preferences, which is too coarse to learn an expressive embedding of current preference. To address the two challenges, we propose a novel model called Time Lag aware Sequential Recommendation (TLSRec), which integrates a hierarchical modeling of user preference and a time lag sensitive fine-grained fusion of the long-term and short-term preferences. TLSRec employs a hierarchical self-attention network to learn users' preference at both global and local time scales, and a neural time gate to adaptively regulate the contributions of the long-term and short-term preferences for the learning of a user's current preference at the aspect level and based on the lag between the current time and the time of the last behavior of a user. The extensive experiments conducted on real datasets verify the effectiveness of TLSRec.

AI Enlightens Wireless Communication: A Transformer Backbone for CSI Feedback

This paper is based on the background of the 2nd Wireless Communication Artificial Intelligence (AI) Competition (WAIC) which is hosted by IMT-2020(5G) Promotion Group 5G+AIWork Group, where the framework of the eigenvector-based channel state information (CSI) feedback problem is firstly provided. Then a basic Transformer backbone for CSI feedback referred to EVCsiNet-T is proposed. Moreover, a series of potential enhancements for deep learning based (DL-based) CSI feedback including i) data augmentation, ii) loss function design, iii) training strategy, and iv) model ensemble are introduced. The experimental results involving the comparison between EVCsiNet-T and traditional codebook methods over different channels are further provided, which show the advanced performance and a promising prospect of Transformer on DL-based CSI feedback problem.