Code-switching (CS), which is when Vietnamese speech uses English words like drug names or procedures, is a common phenomenon in Vietnamese medical communication. This creates challenges for Automatic Speech Recognition (ASR) systems, especially in low-resource languages like Vietnamese. Current most ASR systems struggle to recognize correctly English medical terms within Vietnamese sentences, and no benchmark addresses this challenge. In this paper, we construct a 34-hour Vietnamese Medical Code-Switching Speech dataset (ViMedCSS) containing 16,576 utterances. Each utterance includes at least one English medical term drawn from a curated bilingual lexicon covering five medical topics. Using this dataset, we evaluate several state-of-the-art ASR models and examine different specific fine-tuning strategies for improving medical term recognition to investigate the best approach to solve in the dataset. Experimental results show that Vietnamese-optimized models perform better on general segments, while multilingual pretraining helps capture English insertions. The combination of both approaches yields the best balance between overall and code-switched accuracy. This work provides the first benchmark for Vietnamese medical code-switching and offers insights into effective domain adaptation for low-resource, multilingual ASR systems.

Recent object detectors have achieved impressive accuracy in identifying objects seen during training. However, real-world deployment often introduces novel and unexpected objects, referred to as out-of-distribution (OOD) objects, posing significant challenges to model trustworthiness. Modern object detectors are typically overconfident, making it unreliable to use their predictions alone for OOD detection. To address this, we propose leveraging an auxiliary model as a complementary solution. Specifically, we utilize an off-the-shelf text-to-image generative model, such as Stable Diffusion, which is trained with objective functions distinct from those of discriminative object detectors. We hypothesize that this fundamental difference enables the detection of OOD objects by measuring inconsistencies between the models. Concretely, for a given detected object bounding box and its predicted in-distribution class label, we perform class-conditioned inpainting on the image with the object removed. If the object is OOD, the inpainted image is likely to deviate significantly from the original, making the reconstruction error a robust indicator of OOD status. Extensive experiments demonstrate that our approach consistently surpasses existing zero-shot and non-zero-shot OOD detection methods, establishing a robust framework for enhancing object detection systems in dynamic environments. Our implementation is available at https://github.com/quanghuy0497/RONIN.

Expensive multi-objective optimization problems (EMOPs) are common in real-world scenarios where evaluating objective functions is costly and involves extensive computations or physical experiments. Current Pareto set learning methods for such problems often rely on surrogate models like Gaussian processes to approximate the objective functions. These surrogate models can become fragmented, resulting in numerous small uncertain regions between explored solutions. When using acquisition functions such as the Lower Confidence Bound (LCB), these uncertain regions can turn into pseudo-local optima, complicating the search for globally optimal solutions. To address these challenges, we propose a novel approach called SVH-PSL, which integrates Stein Variational Gradient Descent (SVGD) with Hypernetworks for efficient Pareto set learning. Our method addresses the issues of fragmented surrogate models and pseudo-local optima by collectively moving particles in a manner that smooths out the solution space. The particles interact with each other through a kernel function, which helps maintain diversity and encourages the exploration of underexplored regions. This kernel-based interaction prevents particles from clustering around pseudo-local optima and promotes convergence towards globally optimal solutions. Our approach aims to establish robust relationships between trade-off reference vectors and their corresponding true Pareto solutions, overcoming the limitations of existing methods. Through extensive experiments across both synthetic and real-world MOO benchmarks, we demonstrate that SVH-PSL significantly improves the quality of the learned Pareto set, offering a promising solution for expensive multi-objective optimization problems.

Few-shot image classification remains challenging due to the scarcity of labeled training examples. Augmenting them with synthetic data has emerged as a promising way to alleviate this issue, but models trained on synthetic samples often face performance degradation due to the inherent gap between real and synthetic distributions. To address this limitation, we develop a theoretical framework that quantifies the impact of such distribution discrepancies on supervised learning, specifically in the context of image classification. More importantly, *our framework suggests practical ways to generate good synthetic samples and to train a predictor with high generalization ability*. Building upon this framework, we propose a novel theoretical-based algorithm that integrates prototype learning to optimize both data partitioning and model training, effectively bridging the gap between real few-shot data and synthetic data. Extensive experiments results show that our approach demonstrates superior performance compared to state-of-the-art methods, outperforming them across multiple datasets.

Machine translation for Vietnamese-English in the medical domain is still an under-explored research area. In this paper, we introduce MedEV—a high-quality Vietnamese-English parallel dataset constructed specifically for the medical domain, comprising approximately 360K sentence pairs. We conduct extensive experiments comparing Google Translate, ChatGPT (gpt-3.5-turbo), state-of-the-art Vietnamese-English neural machine translation models and pre-trained bilingual/multilingual sequence-to-sequence models on our new MedEV dataset. Experimental results show that the best performance is achieved by fine-tuning “vinai-translate” for each translation direction. We publicly release our dataset to promote further research.

Conversational Recommender Systems (CRSs) leverage natural language dialogues to provide tailored recommendations.Traditional methods in this field primarily focus on extracting user preferences from isolated dialogues.It often yields responses with a limited perspective, confined to the scope of individual conversations.Recognizing the potential in collective dialogue examples, our research proposes an expanded approach for CRS models, utilizing selective analogues from dialogue histories and responses to enrich both generation and recommendation processes.This introduces significant research challenges, including: (1) How to secure high-quality collections of recommendation dialogue exemplars?(2) How to effectively leverage these exemplars to enhance CRS models?To tackle these challenges, we introduce a novel Demonstrationenhanced Conversational Recommender System (DCRS), which aims to strengthen its understanding on the given dialogue contexts by retrieving and learning from demonstrations.In particular, we first propose a knowledge-aware contrastive learning method that adeptly taps into the mentioned entities and the dialogue's contextual essence for pretraining the demonstration retriever.Subsequently, we further develop two adaptive demonstrationaugmented prompt learning approaches, involving contextualized prompt learning and knowledge-enriched prompt learning, to bridge the gap between the retrieved demonstrations and the two end tasks of CRS, i.e., response generation and item recommendation, respectively.Rigorous evaluations on two established benchmark datasets underscore DCRS's superior performance over existing CRS methods in both item recommendation and response generation 1 .

Federated Recommendation (FedRec) systems have emerged as a solution to safeguard users' data in response to growing regulatory concerns. However, one of the major challenges in these systems lies in the communication costs that arise from the need to transmit neural network models between user devices and a central server. Prior approaches to these challenges often lead to issues such as computational overheads, model specificity constraints, and compatibility issues with secure aggregation protocols. In response, we propose a novel framework, called Correlated Low-rank Structure (CoLR), which leverages the concept of adjusting lightweight trainable parameters while keeping most parameters frozen. Our approach substantially reduces communication overheads without introducing additional computational burdens. Critically, our framework remains fully compatible with secure aggregation protocols, including the robust use of Homomorphic Encryption. The approach resulted in a reduction of up to 93.75% in payload size, with only an approximate 8% decrease in recommendation performance across datasets. Code for reproducing our experiments can be found at https://github.com/NNHieu/CoLR-FedRec.

Pareto Front Learning (PFL) was recently introduced as an effective approach to obtain a mapping function from a given trade-off vector to a solution on the Pareto front, which solves the multi-objective optimization (MOO) problem. Due to the inherent trade-off between conflicting objectives, PFL offers a flexible approach in many scenarios in which the decision makers can not specify the preference of one Pareto solution over another, and must switch between them depending on the situation. However, existing PFL methods ignore the relationship between the solutions during the optimization process, which hinders the quality of the obtained front. To overcome this issue, we propose a novel PFL framework namely PHN-HVI, which employs a hypernetwork to generate multiple solutions from a set of diverse trade-off preferences and enhance the quality of the Pareto front by maximizing the Hypervolume indicator defined by these solutions. The experimental results on several MOO machine learning tasks show that the proposed framework significantly outperforms the baselines in producing the trade-off Pareto front.

The ability to proactively engage with users towards pitching products is highly desired for conversational assistants.However, existing conversational recommendation methods overemphasize on acquiring user preferences while ignore the strategic planning for nudging users towards accepting a designated item.Hence, these methods fail to promote specified items with engaging responses.In this work, we propose a Reinforced Target-driven Conversational Promotion (RTCP) framework for conversational promotion.Specifically, RTCP integrates short-term and long-term planning via a balanced gating mechanism.Inside which, the dialogue strategies are predicted via knowledge-integrated multi-head attention and guided via reinforcement learning rewards.RTCP then employs an action-guided prefix tuning method to generate relevant responses.Experimental results demonstrate that our model outperforms state-of-the-art models on both automatic metrics and human evaluation.Moreover, RTCP has a strong capability in quickly adapting to unseen scenarios just by updating prefix parameters without re-training the whole model.Code and data are here 1 .

Pairwise dot product-based self-attention is key to the success of transformers which achieve state-of-the-art performance across a variety of applications in language and vision, but are costly to compute. It has been shown that most attention scores and keys in transformers are redundant and can be removed without loss of accuracy. In this paper, we develop a novel probabilistic framework for pruning attention scores and keys in transformers. We first formulate an admixture model of attention keys whose input data to be clustered are attention queries. We show that attention scores in self-attention correspond to the posterior distribution of this model when attention keys admit a uniform prior distribution. We then relax this uniform prior constraint and let the model learn these priors from data, resulting in a new Finite Admixture of Keys (FiAK). The learned priors are used for pruning away redundant attention scores and keys in the baseline transformers, improving the diversity of attention patterns that the models capture. We corroborate the efficiency of transformers pruned with FiAK on the ImageNet object classification and WikiText-103 language modeling tasks. Our experiments demonstrate that transformers pruned with FiAK yield similar or better accuracy than the baseline dense transformers while being much more efficient in terms of memory and computational cost.

Multi-head attention is a driving force behind state-of-the-art transformers, which achieve remarkable performance across a variety of natural language processing (NLP) and computer vision tasks. It has been observed that for many applications, those attention heads learn redundant embedding, and most of them can be removed without degrading the performance of the model. Inspired by this observation, we propose Transformer with a Mixture of Gaussian Keys (Transformer-MGK), a novel transformer architecture that replaces redundant heads in transformers with a mixture of keys at each head. These mixtures of keys follow a Gaussian mixture model and allow each attention head to focus on different parts of the input sequence efficiently. Compared to its conventional transformer counterpart, Transformer-MGK accelerates training and inference, has fewer parameters, and requires fewer FLOPs to compute while achieving comparable or better accuracy across tasks. Transformer-MGK can also be easily extended to use with linear attention. We empirically demonstrate the advantage of Transformer-MGK in a range of practical applications, including language modeling and tasks that involve very long sequences. On the Wikitext-103 and Long Range Arena benchmark, Transformer-MGKs with 4 heads attain comparable or better performance to the baseline transformers with 8 heads.

In many visually-oriented applications, users can select and group images that they find interesting into coherent clusters. For instance, we encounter these in the form of hashtags on Instagram, galleries on Flickr, or boards on Pinterest. The selection and coherence of such user-curated visual clusters arise from a user's preference for a certain type of content as well as her own perception of which images are similar and thus belong to a cluster. We seek to model such curation behaviors towards supporting users in their future activities such as expanding existing clusters or discovering new clusters altogether. This paper proposes a framework, namely COLLABORATIVE CURATING that jointly models the interrelated modalities of preference expression and similarity perception. Extensive experiments on real-world datasets from a visual curating platform show that the proposed framework significantly outperforms baselines focusing on either clustering behaviors or preferences alone.

Top-k recommendation seeks to deliver a personalized list of k items to each individual user. An established methodology in the literature based on matrix factorization (MF), which usually represents users and items as vectors in low-dimensional space, is an effective approach to recommender systems, thanks to its superior performance in terms of recommendation quality and scalability. A typical matrix factorization recommender system has two main phases: preference elicitation and recommendation retrieval. The former analyzes user-generated data to learn user preferences and item characteristics in the form of latent feature vectors, whereas the latter ranks the candidate items based on the learnt vectors and returns the top-k items from the ranked list. For preference elicitation, there have been numerous works to build accurate MF-based recommendation algorithms that can learn from large datasets. However, for the recommendation retrieval phase, naively scanning a large number of items to identify the few most relevant ones may inhibit truly real-time applications. In this work, we survey recent advances and state-of-the-art approaches in the literature that enable fast and accurate retrieval for MF-based personalized recommendations. Also, we include analytical discussions of approaches along different dimensions to provide the readers with a more comprehensive understanding of the surveyed works.

Locality Sensitive Hashing (LSH) has become one of the most commonly used approximate nearest neighbor search techniques to avoid the prohibitive cost of scanning through all data points. For recommender systems, LSH achieves efficient recommendation retrieval by encoding user and item vectors into binary hash codes, reducing the cost of exhaustively examining all the item vectors to identify the top-k items. However, conventional matrix factorization models may suffer from performance degeneration caused by randomly-drawn LSH hash functions, directly affecting the ultimate quality of the recommendations. In this paper, we propose a framework named øurmodel, which factors in the stochasticity of LSH hash functions when learning real-valued user and item latent vectors, eventually improving the recommendation accuracy after LSH indexing. Experiments on publicly available datasets show that the proposed framework not only effectively learns user's preferences for prediction, but also achieves high compatibility with LSH stochasticity, producing superior post-LSH indexing performances as compared to state-of-the-art baselines.

Ordinal embedding seeks a low-dimensional representation of objects based on relative comparisons of their similarities. This low-dimensional representation lends itself to visualization on a Euclidean map. Classical assumptions admit only one valid aspect of similarity. However, there are increasing scenarios involving ordinal comparisons that inherently reflect multiple aspects of similarity, which would be better represented by multiple maps. We formulate this problem as conditional ordinal embedding, which learns a distinct low-dimensional representation conditioned on each aspect, yet allows collaboration across aspects via a shared representation. Our geometric approach is novel in its use of a shared spherical representation and multiple aspect-specific projection maps on tangent hyperplanes. Experiments on public datasets showcase the utility of collaborative learning over baselines that learn multiple maps independently.

Determining the similarity between two objects is pertinent to many applications. When the basis for similarity is a set of object-to-object relationships, it is natural to rely on graph-theoretic measures. One seminal technique for measuring the structural-context similarity between a pair of graph vertices is SimRank, whose underlying intuition is that two objects are similar if they are connected by similar objects. However, by design, SimRank as well as its variants capture only a single view or perspective of similarity. Meanwhile, in many real-world scenarios, there emerge multiple perspectives of similarity, i.e., two objects may be similar from one perspective, but dissimilar from another. For instance, human subjects may generate varied, yet valid, clusterings of objects. In this work, we propose a graph-theoretic similarity measure that is natively multiperspective. In our approach, the observed object-to-object relationships due to various perspectives are integrated into a unified graph-based representation, stylised as a hypergraph to retain the distinct perspectives. We then introduce a novel model for learning and reflecting diverse similarity perceptions given the hypergraph, yielding the similarity score between any pair of objects from any perspective. In addition to proposing an algorithm for computing the similarity scores, we also provide theoretical guarantees on the convergence of the algorithm. Experiments on public datasets show that the proposed model deals better with multiperspectivity than the baselines.

Top-k recommendation seeks to deliver a personalized recommendation list of k items to a user. The dual objectives are (1) accuracy in identifying the items a user is likely to prefer, and (2) efficiency in constructing the recommendation list in real time. One direction towards retrieval efficiency is to formulate retrieval as approximate k nearest neighbor (kNN) search aided by indexing schemes, such as locality-sensitive hashing, spatial trees, and inverted index. These schemes, applied on the output representations of recommendation algorithms, speed up the retrieval process by automatically discarding a large number of potentially irrelevant items when given a user query vector. However, many previous recommendation algorithms produce representations that may not necessarily align well with the structural properties of these indexing schemes, eventually resulting in a significant loss of accuracy post-indexing. In this paper, we introduce Indexable Bayesian Personalized Ranking (IBPR) that learns from ordinal preference to produce representation that is inherently compatible with the aforesaid indices. Experiments on publicly available datasets show superior performance of the proposed model compared to state-of-the-art methods on top-k recommendation retrieval task, achieving significant speedup while maintaining high accuracy.

Embedding deals with reducing the high-dimensional representation of data into a low-dimensional representation. Previous work mostly focuses on preserving similarities among objects. Here, not only do we explicitly recognize multiple types of objects, but we also focus on the ordinal relationships across types. Collaborative Ordinal Embedding or COE is based on generative modelling of ordinal triples. Experiments show that COE outperforms the baselines on objective metrics, revealing its capacity for information preservation for ordinal data.