Different languages have distinct phonetic systems and vary in their prosodic features making it challenging to develop a Text-to-Speech (TTS) model that can effectively synthesise speech in multilingual settings. Furthermore, TTS architecture needs to be both efficient enough to capture nuances in multiple languages and efficient enough to be practical for deployment. The standard approach is to build transformer based model such as SpeechT5 and train it on large multilingual dataset. As the size of these models grow the conventional fine-tuning for adapting these model becomes impractical due to heavy computational cost. In this paper, we proposes to integrate parameter-efficient transfer learning (PETL) methods such as adapters and hypernetwork with TTS architecture for multilingual speech synthesis. Notably, in our experiments PETL methods able to achieve comparable or even better performance compared to full fine-tuning with only $\sim$2.5\% tunable parameters.The code and samples are available at: https://anonymous.4open.science/r/multilingualTTS-BA4C.
The expansion of model parameters underscores the significance of pre-trained models; however, the constraints encountered during model deployment necessitate models of variable sizes. Consequently, the traditional pre-training and fine-tuning paradigm fails to address the initialization problem when target models are incompatible with pre-trained models. We tackle this issue from a multitasking perspective and introduce \textbf{WAVE}, which incorporates a set of shared \textbf{W}eight templates for \textbf{A}daptive initialization of \textbf{V}ariable-siz\textbf{E}d Models. During initialization, target models will initialize the corresponding weight scalers tailored to their model size, which are sufficient to learn the connection rules of weight templates based on the Kronecker product from a limited amount of data. For the construction of the weight templates, WAVE utilizes the \textit{Learngene} framework, which structurally condenses common knowledge from ancestry models into weight templates as the learngenes through knowledge distillation. This process allows the integration of pre-trained models' knowledge into structured knowledge according to the rules of weight templates. We provide a comprehensive benchmark for the learngenes, and extensive experiments demonstrate the efficacy of WAVE. The results show that WAVE achieves state-of-the-art performance when initializing models with various depth and width, and even outperforms the direct pre-training of $n$ entire models, particularly for smaller models, saving approximately $n\times$ and $5\times$ in computational and storage resources, respectively. WAVE simultaneously achieves the most efficient knowledge transfer across a series of datasets, specifically achieving an average improvement of 1.8\% and 1.2\% on 7 downstream datasets.
Cross-Domain Recommendation (CDR) seeks to utilize knowledge from different domains to alleviate the problem of data sparsity in the target recommendation domain, and it has been gaining more attention in recent years. Although there have been notable advancements in this area, most current methods represent users and items in Euclidean space, which is not ideal for handling long-tail distributed data in recommendation systems. Additionally, adding data from other domains can worsen the long-tail characteristics of the entire dataset, making it harder to train CDR models effectively. Recent studies have shown that hyperbolic methods are particularly suitable for modeling long-tail distributions, which has led us to explore hyperbolic representations for users and items in CDR scenarios. However, due to the distinct characteristics of the different domains, applying hyperbolic representation learning to CDR tasks is quite challenging. In this paper, we introduce a new framework called Hyperbolic Contrastive Learning (HCTS), designed to capture the unique features of each domain while enabling efficient knowledge transfer between domains. We achieve this by embedding users and items from each domain separately and mapping them onto distinct hyperbolic manifolds with adjustable curvatures for prediction. To improve the representations of users and items in the target domain, we develop a hyperbolic contrastive learning module for knowledge transfer. Extensive experiments on real-world datasets demonstrate that hyperbolic manifolds are a promising alternative to Euclidean space for CDR tasks.
Large language models (LLMs) possess extensive parametric knowledge, but this knowledge is difficult to update with new information because retraining is very expensive and infeasible for closed-source models. Knowledge editing (KE) has emerged as a viable solution for updating the knowledge of LLMs without compromising their overall performance. On-the-fly KE methods, inspired by in-context learning (ICL), have shown great promise and allow LLMs to be treated as black boxes. In the past, KE was primarily employed in English contexts, whereas the potential for cross-lingual KE in current English-centric LLMs has not been fully explored. To foster more research in this direction, we introduce the BMIKE-53 benchmark for evaluating cross-lingual KE on 53 diverse languages across three KE task types. We also propose a gradient-free KE method called Multilingual In-context Knowledge Editing (MIKE) and evaluate it on BMIKE-53. Our evaluation focuses on cross-lingual knowledge transfer in terms of reliability, generality, locality, and portability, offering valuable insights and a framework for future research in cross-lingual KE. Our code and data are publicly accessible via the anonymous repository at https://anonymous.4open.science/r/MIKE.
Ensuring smooth mobility management while employing directional beamformed transmissions in 5G millimeter-wave networks calls for robust and accurate user equipment (UE) localization and tracking. In this article, we develop neural network-based positioning models with time- and frequency-domain channel state information (CSI) data in harsh non-line-of-sight (NLoS) conditions. We propose a novel frequency-domain feature extraction, which combines relative phase differences and received powers across resource blocks, and offers robust performance and reliability. Additionally, we exploit the multipath components and propose an aggregate time-domain feature combining time-of-flight, angle-of-arrival and received path-wise powers. Importantly, the temporal correlations are also harnessed in the form of sequence processing neural networks, which prove to be of particular benefit for vehicular UEs. Realistic numerical evaluations in large-scale line-of-sight (LoS)-obstructed urban environment with moving vehicles are provided, building on full ray-tracing based propagation modeling. The results show the robustness of the proposed CSI features in terms of positioning accuracy, and that the proposed models reliably localize UEs even in the absence of a LoS path, clearly outperforming the state-of-the-art with similar or even reduced processing complexity. The proposed sequence-based neural network model is capable of tracking the UE position, speed and heading simultaneously despite the strong uncertainties in the CSI measurements. Finally, it is shown that differences between the training and online inference environments can be efficiently addressed and alleviated through transfer learning.
This study delves into the classification of interference signals to global navigation satellite systems (GNSS) stemming from mobile jammers such as unmanned aerial vehicles (UAVs) across diverse wireless communication zones, employing federated learning (FL) and transfer learning (TL). Specifically, we employ a neural network classifier, enhanced with FL to decentralize data processing and TL to hasten the training process, aiming to improve interference classification accuracy while preserving data privacy. Our evaluations span multiple data scenarios, incorporating both independent and identically distributed (IID) and non-identically distributed (non-IID), to gauge the performance of our approach under different interference conditions. Our results indicate an improvement of approximately $8\%$ in classification accuracy compared to basic convolutional neural network (CNN) model, accompanied by expedited convergence in networks utilizing pre-trained models. Additionally, the implementation of FL not only developed privacy but also matched the robustness of centralized learning methods, particularly under IID scenarios. Moreover, the federated averaging (FedAvg) algorithm effectively manages regional interference variability, thereby enhancing the regional communication performance indicator, $C/N_0$, by roughly $5\text{dB}\cdot \text{Hz}$ compared to isolated setups.
This paper tackles the task of emotion-cause pair extraction in the unsupervised domain adaptation setting. The problem is challenging as the distributions of the events causing emotions in target domains are dramatically different than those in source domains, despite the distributions of emotional expressions between domains are overlapped. Inspired by causal discovery, we propose a novel deep latent model in the variational autoencoder (VAE) framework, which not only captures the underlying latent structures of data but also utilizes the easily transferable knowledge of emotions as the bridge to link the distributions of events in different domains. To facilitate knowledge transfer across domains, we also propose a novel variational posterior regularization technique to disentangle the latent representations of emotions from those of events in order to mitigate the damage caused by the spurious correlations related to the events in source domains. Through extensive experiments, we demonstrate that our model outperforms the strongest baseline by approximately 11.05% on a Chinese benchmark and 2.45% on a English benchmark in terms of weighted-average F1 score. The source code will be publicly available upon acceptance.
Efficient fine-tuning of large language models for task-specific applications is imperative, yet the vast number of parameters in these models makes their training increasingly challenging. Despite numerous proposals for effective methods, a substantial memory overhead remains for gradient computations during updates. \thm{Can we fine-tune a series of task-specific small models and transfer their knowledge directly to a much larger model without additional training?} In this paper, we explore weak-to-strong specialization using logit arithmetic, facilitating a direct answer to this question. Existing weak-to-strong methods often employ a static knowledge transfer ratio and a single small model for transferring complex knowledge, which leads to suboptimal performance. % To address this, To surmount these limitations, we propose a dynamic logit fusion approach that works with a series of task-specific small models, each specialized in a different task. This method adaptively allocates weights among these models at each decoding step, learning the weights through Kullback-Leibler divergence constrained optimization problems. We conduct extensive experiments across various benchmarks in both single-task and multi-task settings, achieving leading results. By transferring expertise from the 7B model to the 13B model, our method closes the performance gap by 96.4\% in single-task scenarios and by 86.3\% in multi-task scenarios compared to full fine-tuning of the 13B model. Notably, we achieve surpassing performance on unseen tasks. Moreover, we further demonstrate that our method can effortlessly integrate in-context learning for single tasks and task arithmetic for multi-task scenarios.
Recently, there has been a growing availability of pre-trained text models on various model repositories. These models greatly reduce the cost of training new models from scratch as they can be fine-tuned for specific tasks or trained on large datasets. However, these datasets may not be publicly accessible due to the privacy, security, or intellectual property issues. In this paper, we aim to develop a lightweight student network that can learn from multiple teacher models without accessing their original training data. Hence, we investigate Data-Free Knowledge Amalgamation (DFKA), a knowledge-transfer task that combines insights from multiple pre-trained teacher models and transfers them effectively to a compact student network. To accomplish this, we propose STRATANET, a modeling framework comprising: (a) a steerable data generator that produces text data tailored to each teacher and (b) an amalgamation module that implements a self-regulative strategy using confidence estimates from the teachers' different layers to selectively integrate their knowledge and train a versatile student. We evaluate our method on three benchmark text classification datasets with varying labels or domains. Empirically, we demonstrate that the student model learned using our STRATANET outperforms several baselines significantly under data-driven and data-free constraints.
Large language models (LLMs) are typically multilingual due to pretraining on diverse multilingual corpora. But can these models relate corresponding concepts across languages, effectively being crosslingual? This study evaluates six state-of-the-art LLMs on inherently crosslingual tasks. We observe that while these models show promising surface-level crosslingual abilities on machine translation and embedding space analyses, they struggle with deeper crosslingual knowledge transfer, revealing a crosslingual knowledge barrier in both general (MMLU benchmark) and domain-specific (Harry Potter quiz) contexts. We observe that simple inference-time mitigation methods offer only limited improvement. On the other hand, we propose fine-tuning of LLMs on mixed-language data, which effectively reduces these gaps, even when using out-of-domain datasets like WikiText. Our findings suggest the need for explicit optimization to unlock the full crosslingual potential of LLMs. Our code is publicly available at https://github.com/google-research/crosslingual-knowledge-barriers.
Bayesian neural network (BNN) approximates the posterior distribution of model parameters and utilizes the posterior for prediction via Bayesian Model Averaging (BMA). The quality of the posterior approximation is critical for achieving accurate and robust predictions. It is known that flatness in the loss landscape is strongly associated with generalization performance, and it necessitates consideration to improve the quality of the posterior approximation. In this work, we empirically demonstrate that BNNs often struggle to capture the flatness. Moreover, we provide both experimental and theoretical evidence showing that BMA can be ineffective without ensuring flatness. To address this, we propose Sharpness-Aware Bayesian Model Averaging (SA-BMA), a novel optimizer that seeks flat posteriors by calculating divergence in the parameter space. SA-BMA aligns with the intrinsic nature of BNN and the generalized version of existing sharpness-aware optimizers for DNN. In addition, we suggest a Bayesian Transfer Learning scheme to efficiently leverage pre-trained DNN. We validate the efficacy of SA-BMA in enhancing generalization performance in few-shot classification and distribution shift by ensuring flat posterior.
Experiments in affective computing are based on stimulus datasets that, in the process of standardization, receive metadata describing which emotions each stimulus evokes. In this paper, we explore an approach to creating stimulus datasets for affective computing using generative adversarial networks (GANs). Traditional dataset preparation methods are costly and time consuming, prompting our investigation of alternatives. We conducted experiments with various GAN architectures, including Deep Convolutional GAN, Conditional GAN, Auxiliary Classifier GAN, Progressive Augmentation GAN, and Wasserstein GAN, alongside data augmentation and transfer learning techniques. Our findings highlight promising advances in the generation of emotionally evocative synthetic images, suggesting significant potential for future research and improvements in this domain.
The manual examination of X-ray images for fractures is a time-consuming process that is prone to human error. In this work, we introduce a robust yet simple training loop for the classification of fractures, which significantly outperforms existing methods. Our method achieves superior performance in less than ten epochs and utilizes the latest dataset to deliver the best-performing model for this task. We emphasize the importance of training deep learning models responsibly and efficiently, as well as the critical role of selecting high-quality datasets.
As a crucial step toward real-world learning scenarios with changing environments, dataset shift theory and invariant representation learning algorithm have been extensively studied to relax the identical distribution assumption in classical learning setting. Among the different assumptions on the essential of shifting distributions, generalized label shift (GLS) is the latest developed one which shows great potential to deal with the complex factors within the shift. In this paper, we aim to explore the limitations of current dataset shift theory and algorithm, and further provide new insights by presenting a comprehensive understanding of GLS. From theoretical aspect, two informative generalization bounds are derived, and the GLS learner is proved to be sufficiently close to optimal target model from the Bayesian perspective. The main results show the insufficiency of invariant representation learning, and prove the sufficiency and necessity of GLS correction for generalization, which provide theoretical supports and innovations for exploring generalizable model under dataset shift. From methodological aspect, we provide a unified view of existing shift correction frameworks, and propose a kernel embedding-based correction algorithm (KECA) to minimize the generalization error and achieve successful knowledge transfer. Both theoretical results and extensive experiment evaluations demonstrate the sufficiency and necessity of GLS correction for addressing dataset shift and the superiority of proposed algorithm.
DESI is a groundbreaking international project to observe more than 40 million quasars and galaxies over a 5-year period to create a 3D map of the sky. This map will enable us to probe multiple aspects of cosmology, from dark energy to neutrino mass. We are focusing here on one type of object observed by DESI, the Lyman Break Galaxies (LBGs). The aim is to use their spectra to determine whether they are indeed LBGs, and if so, to determine their distance from the Earth using a phenomenon called redshift. This will enable us to place these galaxies on the DESI 3D map. The aim is therefore to develop a convolutional neural network (CNN) inspired by QuasarNET (See arXiv:1808.09955), performing simultaneously a classification (LBG type or not) and a regression task (determine the redshift of the LBGs). Initially, data augmentation techniques such as shifting the spectra in wavelengths, adding noise to the spectra, or adding synthetic spectra were used to increase the model training dataset from 3,019 data to over 66,000. In a second phase, modifications to the QuasarNET architecture, notably through transfer learning and hyperparameter tuning with Bayesian optimization, boosted model performance. Gains of up to 26% were achieved on the Purity/Efficiency curve, which is used to evaluate model performance, particularly in areas with interesting redshifts, at low (around 2) and high (around 4) redshifts. The best model obtained an average score of 94%, compared with 75% for the initial model.
This paper introduces a novel framework for matrix diagonalization, recasting it as a sequential decision-making problem and applying the power of Decision Transformers (DTs). Our approach determines optimal pivot selection during diagonalization with the Jacobi algorithm, leading to significant speedups compared to the traditional max-element Jacobi method. To bolster robustness, we integrate an epsilon-greedy strategy, enabling success in scenarios where deterministic approaches fail. This work demonstrates the effectiveness of DTs in complex computational tasks and highlights the potential of reimagining mathematical operations through a machine learning lens. Furthermore, we establish the generalizability of our method by using transfer learning to diagonalize matrices of smaller sizes than those trained.
Multi-Task Evolutionary Optimization (MTEO), an important field focusing on addressing complex problems through optimizing multiple tasks simultaneously, has attracted much attention. While MTEO has been primarily focusing on task similarity, there remains a hugely untapped potential in harnessing the shared characteristics between different domains to enhance evolutionary optimization. For example, real-world complex systems usually share the same characteristics, such as the power-law rule, small-world property, and community structure, thus making it possible to transfer solutions optimized in one system to another to facilitate the optimization. Drawing inspiration from this observation of shared characteristics within complex systems, we set out to extend MTEO to a novel framework - multi-domain evolutionary optimization (MDEO). To examine the performance of the proposed MDEO, we utilize a challenging combinatorial problem of great security concern - community deception in complex networks as the optimization task. To achieve MDEO, we propose a community-based measurement of graph similarity to manage the knowledge transfer among domains. Furthermore, we develop a graph representation-based network alignment model that serves as the conduit for effectively transferring solutions between different domains. Moreover, we devise a self-adaptive mechanism to determine the number of transferred solutions from different domains and introduce a novel mutation operator based on the learned mapping to facilitate the utilization of knowledge from other domains. Experiments on eight real-world networks of different domains demonstrate MDEO superiority in efficacy compared to classical evolutionary optimization. Simulations of attacks on the community validate the effectiveness of the proposed MDEO in safeguarding community security.
Recently, outstanding identification rates in image classification tasks were achieved by convolutional neural networks (CNNs). to use such skills, selective CNNs trained on a dataset of well-known images of metal surface defects captured with an RGB camera. Defects must be detected early to take timely corrective action due to production concerns. For image classification up till now, a model-based method has been utilized, which indicated the predicted reflection characteristics of surface defects in comparison to flaw-free surfaces. The problem of detecting steel surface defects has grown in importance as a result of the vast range of steel applications in end-product sectors such as automobiles, households, construction, etc. The manual processes for detections are time-consuming, labor-intensive, and expensive. Different strategies have been used to automate manual processes, but CNN models have proven to be the most effective rather than image processing and machine learning techniques. By using different CNN models with fine-tuning, easily compare their performance and select the best-performing model for the same kinds of tasks. However, it is important that using different CNN models either from fine tuning can be computationally expensive and time-consuming. Therefore, our study helps the upcoming researchers to choose the CNN without considering the issues of model complexity, performance, and computational resources. In this article, the performance of various CNN models with transfer learning techniques are evaluated. These models were chosen based on their popularity and impact in the field of computer vision research, as well as their performance on benchmark datasets. According to the outcomes, DenseNet201 outperformed the other CNN models and had the greatest detection rate on the NEU dataset, falling in at 98.37 percent.
Machine Learning-based heuristics have recently shown impressive performance in solving a variety of hard combinatorial optimization problems (COPs). However they generally rely on a separate neural model, specialized and trained for each single problem. Any variation of a problem requires adjustment of its model and re-training from scratch. In this paper, we propose GOAL (for Generalist combinatorial Optimization Agent Learning), a generalist model capable of efficiently solving multiple COPs and which can be fine-tuned to solve new COPs. GOAL consists of a single backbone plus light-weight problem-specific adapters for input and output processing. The backbone is based on a new form of mixed-attention blocks which allows to handle problems defined on graphs with arbitrary combinations of node, edge and instance-level features. Additionally, problems which involve heterogeneous types of nodes or edges are handled through a novel multi-type transformer architecture, where the attention blocks are duplicated to attend the meaningful combinations of types while relying on the same shared parameters. We train GOAL on a set of routing, scheduling and classic graph problems and show that it is only slightly inferior to the specialized baselines while being the first multi-task model that solves a wide range of COPs. Finally we showcase the strong transfer learning capacity of GOAL by fine-tuning it on several new problems. Our code is available at https://github.com/naver/goal-co/.
Recent evaluations of cross-domain text classification models aim to measure the ability of a model to obtain domain-invariant performance in a target domain given labeled samples in a source domain. The primary strategy for this evaluation relies on assumed differences between source domain samples and target domain samples in benchmark datasets. This evaluation strategy fails to account for the similarity between source and target domains, and may mask when models fail to transfer learning to specific target samples which are highly dissimilar from the source domain. We introduce Depth $F_1$, a novel cross-domain text classification performance metric. Designed to be complementary to existing classification metrics such as $F_1$, Depth $F_1$ measures how well a model performs on target samples which are dissimilar from the source domain. We motivate this metric using standard cross-domain text classification datasets and benchmark several recent cross-domain text classification models, with the goal of enabling in-depth evaluation of the semantic generalizability of cross-domain text classification models.