Speaker diarization answers the question "who spoke when" for an audio file. In some diarization scenarios, low latency is required for transcription. Speaker diarization with low latency is referred to as online speaker diarization. The DIART pipeline is an online speaker diarization system. It consists of a segmentation and an embedding model. The embedding model has the largest share of the overall latency. The aim of this paper is to optimize the inference latency of the DIART pipeline. Different inference optimization methods such as knowledge distilation, pruning, quantization and layer fusion are applied to the embedding model of the pipeline. It turns out that knowledge distillation optimizes the latency, but has a negative effect on the accuracy. Quantization and layer fusion also have a positive influence on the latency without worsening the accuracy. Pruning, on the other hand, does not improve latency.
The e-commerce platform has evolved rapidly due to its widespread popularity and convenience. Developing an e-commerce shopping assistant for customers is crucial to aiding them in quickly finding desired products and recommending precisely what they need. However, most previous shopping assistants face two main problems: (1) task-specificity, which necessitates the development of different models for various tasks, thereby increasing development costs and limiting effectiveness; and (2) poor generalization, where the trained model performs inadequately on up-to-date products. To resolve these issues, we employ Large Language Models (LLMs) to construct an omnipotent assistant, leveraging their adeptness at handling multiple tasks and their superior generalization capability. Nonetheless, LLMs lack inherent knowledge of e-commerce concepts. To address this, we create an instruction dataset comprising 65,000 samples and diverse tasks, termed as EshopInstruct. Through instruction tuning on our dataset, the assistant, named LLaSA, demonstrates the potential to function as an omnipotent assistant. Additionally, we propose various inference optimization strategies to enhance performance with limited inference resources. In the Amazon KDD Cup 2024 Challenge, our proposed method, LLaSA, achieved an overall ranking of 3rd place on ShopBench, including 57 tasks and approximately 20,000 questions, and we secured top-5 rankings in each track, especially in track4, where we achieved the best performance result among all student teams. Our extensive practices fully demonstrate that LLMs possess the great potential to be competent e-commerce shopping assistants.
This paper introduces Patched MOA (Mixture of Agents), an inference optimization technique that significantly enhances the performance of large language models (LLMs) across diverse software development tasks. We evaluate three inference optimization algorithms - Best of N, Mixture of Agents, and Monte Carlo Tree Search and demonstrate that Patched MOA can boost the performance of smaller models to surpass that of larger, more expensive models. Notably, our approach improves the gpt-4o-mini model's performance on the Arena-Hard-Auto benchmark by 15.52%, outperforming gpt-4-turbo at a fraction of the cost. We also apply Patched MOA to various software development workflows, showing consistent improvements in task completion rates. Our method is model-agnostic, transparent to end-users, and can be easily integrated into existing LLM pipelines. This work contributes to the growing field of LLM optimization, offering a cost-effective solution for enhancing model performance without the need for fine-tuning or larger models. Our implementation is open-source and available at https://github.com/codelion/optillm.
Powerful foundation models, including large language models (LLMs), with Transformer architectures have ushered in a new era of Generative AI across various industries. Industry and research community have witnessed a large number of new applications, based on those foundation models. Such applications include question and answer, customer services, image and video generation, and code completions, among others. However, as the number of model parameters reaches to hundreds of billions, their deployment incurs prohibitive inference costs and high latency in real-world scenarios. As a result, the demand for cost-effective and fast inference using AI accelerators is ever more higher. To this end, our tutorial offers a comprehensive discussion on complementary inference optimization techniques using AI accelerators. Beginning with an overview of basic Transformer architectures and deep learning system frameworks, we deep dive into system optimization techniques for fast and memory-efficient attention computations and discuss how they can be implemented efficiently on AI accelerators. Next, we describe architectural elements that are key for fast transformer inference. Finally, we examine various model compression and fast decoding strategies in the same context.
Large language models (LLMs) have shown exceptional performance and vast potential across diverse tasks. However, the deployment of LLMs with high performance in low-resource environments has garnered significant attention in the industry. When GPU hardware resources are limited, we can explore alternative options on CPUs. To mitigate the financial burden and alleviate constraints imposed by hardware resources, optimizing inference performance is necessary. In this paper, we introduce an easily deployable inference performance optimization solution aimed at accelerating LLMs on CPUs. In this solution, we implement an effective way to reduce the KV cache size while ensuring precision. We propose a distributed inference optimization approach and implement it based on oneAPI Collective Communications Library. Furthermore, we propose optimization approaches for LLMs on CPU, and conduct tailored optimizations for the most commonly used models. The code is open-sourced at https://github.com/intel/xFasterTransformer.
In this paper, we propose reverse inference optimization (RIO), a simple and effective method designed to enhance the robustness of autoregressive-model-based zero-shot text-to-speech (TTS) systems using reinforcement learning from human feedback (RLHF). To assess the quality of speech produced by the TTS system without human annotations, RIO introduces a novel concept termed as reverse inference based on the Bayesian principle, which suggests that a high-quality generated speech should be able to be used as a prompt for subsequent generation using the same TTS model. By leveraging reverse inference as the standard to select exemplars used in RLHF from the speech samples generated by the TTS system itself, RIO steers the subsequent optimization towards a direction of enhancing the TTS robustness. The RIO framework, comprising sampling, automatic annotating, and learning, obviates the need for a reward model or pairwise preference data, and significantly improves the stability of zero-shot TTS performance by reducing the discrepancies between training and inference conditions. Our experimental results verify that RIO can effectively improve both subjective and objective metrics, including mean opinion scores, word error rates, and speaker similarity. Remarkably, RIO can also diminish the incidence of bad outputs to nearly zero percent, rivalling the robustness when using ground-truth speech as the prompt.
Generative Retrieval introduces a new approach to Information Retrieval by reframing it as a constrained generation task, leveraging recent advancements in Autoregressive (AR) language models. However, AR-based Generative Retrieval methods suffer from high inference latency and cost compared to traditional dense retrieval techniques, limiting their practical applicability. This paper investigates fully Non-autoregressive (NAR) language models as a more efficient alternative for generative retrieval. While standard NAR models alleviate latency and cost concerns, they exhibit a significant drop in retrieval performance (compared to AR models) due to their inability to capture dependencies between target tokens. To address this, we question the conventional choice of limiting the target token space to solely words or sub-words. We propose PIXAR, a novel approach that expands the target vocabulary of NAR models to include multi-word entities and common phrases (up to 5 million tokens), thereby reducing token dependencies. PIXAR employs inference optimization strategies to maintain low inference latency despite the significantly larger vocabulary. Our results demonstrate that PIXAR achieves a relative improvement of 31.0% in MRR@10 on MS MARCO and 23.2% in Hits@5 on Natural Questions compared to standard NAR models with similar latency and cost. Furthermore, online A/B experiments on a large commercial search engine show that PIXAR increases ad clicks by 5.08% and revenue by 4.02%.
The internal structure and operation mechanism of large-scale language models are analyzed theoretically, especially how Transformer and its derivative architectures can restrict computing efficiency while capturing long-term dependencies. Further, we dig deep into the efficiency bottleneck of the training phase, and evaluate in detail the contribution of adaptive optimization algorithms (such as AdamW), massively parallel computing techniques, and mixed precision training strategies to accelerate convergence and reduce memory footprint. By analyzing the mathematical principles and implementation details of these algorithms, we reveal how they effectively improve training efficiency in practice. In terms of model deployment and inference optimization, this paper systematically reviews the latest advances in model compression techniques, focusing on strategies such as quantification, pruning, and knowledge distillation. By comparing the theoretical frameworks of these techniques and their effects in different application scenarios, we demonstrate their ability to significantly reduce model size and inference delay while maintaining model prediction accuracy. In addition, this paper critically examines the limitations of current efficiency optimization methods, such as the increased risk of overfitting, the control of performance loss after compression, and the problem of algorithm generality, and proposes some prospects for future research. In conclusion, this study provides a comprehensive theoretical framework for understanding the efficiency optimization of large-scale language models.
The integration of Foundation Models (FMs) with Federated Learning (FL) presents a transformative paradigm in Artificial Intelligence (AI). This integration offers enhanced capabilities, while addressing concerns of privacy, data decentralization and computational efficiency. This paper provides a comprehensive survey of the emerging field of Federated Foundation Models (FedFM), elucidating their synergistic relationship and exploring novel methodologies, challenges, and future directions that the FL research field needs to focus on in order to thrive in the age of FMs. A systematic multi-tiered taxonomy is proposed, categorizing existing FedFM approaches for model training, aggregation, trustworthiness, and incentivization. Key challenges, including how to enable FL to deal with high complexity of computational demands, privacy considerations, contribution evaluation, and communication efficiency, are thoroughly discussed. Moreover, this paper explores the intricate challenges of communication, scalability and security inherent in training/fine-tuning FMs via FL. It highlights the potential of quantum computing to revolutionize the processes of training, inference, optimization and security. This survey also introduces the implementation requirement of FedFM and some practical FedFM applications. It highlights lessons learned with a clear understanding of our findings for FedFM. Finally, this survey not only provides insights into the current state and challenges of FedFM, but also offers a blueprint for future research directions, emphasizing the need for developing trustworthy solutions. It serves as a foundational guide for researchers and practitioners interested in contributing to this interdisciplinary and rapidly advancing field.
While providing machine learning model as a service to process users' inference requests, online applications can periodically upgrade the model utilizing newly collected data. Federated learning (FL) is beneficial for enabling the training of models across distributed clients while keeping the data locally. However, existing work has overlooked the coexistence of model training and inference under clients' limited resources. This paper focuses on the joint optimization of model training and inference to maximize inference performance at clients. Such an optimization faces several challenges. The first challenge is to characterize the clients' inference performance when clients may partially participate in FL. To resolve this challenge, we introduce a new notion of age of model (AoM) to quantify client-side model freshness, based on which we use FL's global model convergence error as an approximate measure of inference performance. The second challenge is the tight coupling among clients' decisions, including participation probability in FL, model download probability, and service rates. Toward the challenges, we propose an online problem approximation to reduce the problem complexity and optimize the resources to balance the needs of model training and inference. Experimental results demonstrate that the proposed algorithm improves the average inference accuracy by up to 12%.
Advances in machine learning have made it possible to perform various text and speech processing tasks, such as automatic speech recognition (ASR), in an end-to-end (E2E) manner. E2E approaches utilizing pre-trained models are gaining attention for conserving training data and resources. However, most of their applications in ASR involve only one of either a pre-trained speech or a language model. This paper proposes integrating a pre-trained speech representation model and a large language model (LLM) for E2E ASR. The proposed model enables the optimization of the entire ASR process, including acoustic feature extraction and acoustic and language modeling, by combining pre-trained models with a bridge network and also enables the application of remarkable developments in LLM utilization, such as parameter-efficient domain adaptation and inference optimization. Experimental results demonstrate that the proposed model achieves a performance comparable to that of modern E2E ASR models by utilizing powerful pre-training models with the proposed integrated approach.
Recent quantization techniques have enabled heterogeneous precisions at very fine granularity, e.g., each parameter/activation can take on a different precision, resulting in compact neural networks without sacrificing accuracy. However, there is a lack of efficient architectural support for such networks, which require additional hardware to decode the precision settings for individual variables, align the variables, and provide fine-grained mixed-precision compute capabilities. The complexity of these operations introduces high overheads. Thus, the improvements in inference latency/energy of these networks are not commensurate with the compression ratio, and may be inferior to larger quantized networks with uniform precisions. We present an end-to-end co-design approach encompassing computer architecture, training algorithm, and inference optimization to efficiently execute networks with fine-grained heterogeneous precisions. The key to our approach is a novel training algorithm designed to accommodate hardware constraints and inference operation requirements, outputting networks with input-channel-wise heterogeneous precisions and at most three precision levels. Combined with inference optimization techniques, existing architectures with low-cost enhancements can support such networks efficiently, yielding optimized tradeoffs between accuracy, compression ratio and inference latency/energy. We demonstrate the efficacy of our approach across CPU and GPU architectures. For various representative neural networks, our approach achieves >10x improvements in both compression ratio and inference latency, with negligible degradation in accuracy compared to full-precision networks.
In many settings, such as scientific inference, optimization, and transfer learning, the learner has a well-defined objective, which can be treated as estimation of a target parameter, and no intrinsic interest in characterizing the entire data-generating process. Usually, the learner must also contend with additional sources of uncertainty or variables -- with nuisance parameters. Bayesian active learning, or sequential optimal experimental design, can straightforwardly accommodate the presence of nuisance parameters, and so is a natural active learning framework for such problems. However, the introduction of nuisance parameters can lead to bias in the Bayesian learner's estimate of the target parameters, a phenomenon we refer to as negative interference. We characterize the threat of negative interference and how it fundamentally changes the nature of the Bayesian active learner's task. We show that the extent of negative interference can be extremely large, and that accurate estimation of the nuisance parameters is critical to reducing it. The Bayesian active learner is confronted with a dilemma: whether to spend a finite acquisition budget in pursuit of estimation of the target or of the nuisance parameters. Our setting encompasses Bayesian transfer learning as a special case, and our results shed light on the phenomenon of negative transfer between learning environments.
Obtaining sparse, interpretable representations of observable data is crucial in many machine learning and signal processing tasks. For data representing flows along the edges of a graph, an intuitively interpretable way to obtain such representations is to lift the graph structure to a simplicial complex: The eigenvectors of the associated Hodge-Laplacian, respectively the incidence matrices of the corresponding simplicial complex then induce a Hodge decomposition, which can be used to represent the observed data in terms of gradient, curl, and harmonic flows. In this paper, we generalize this approach to cellular complexes and introduce the flow representation learning problem, i.e., the problem of augmenting the observed graph by a set of cells, such that the eigenvectors of the associated Hodge Laplacian provide a sparse, interpretable representation of the observed edge flows on the graph. We show that this problem is NP-hard and introduce an efficient approximation algorithm for its solution. Experiments on real-world and synthetic data demonstrate that our algorithm outperforms state-of-the-art methods with respect to approximation error, while being computationally efficient.
Painterly image harmonization aims to insert photographic objects into paintings and obtain artistically coherent composite images. Previous methods for this task mainly rely on inference optimization or generative adversarial network, but they are either very time-consuming or struggling at fine control of the foreground objects (e.g., texture and content details). To address these issues, we propose a novel Painterly Harmonization stable Diffusion model (PHDiffusion), which includes a lightweight adaptive encoder and a Dual Encoder Fusion (DEF) module. Specifically, the adaptive encoder and the DEF module first stylize foreground features within each encoder. Then, the stylized foreground features from both encoders are combined to guide the harmonization process. During training, besides the noise loss in diffusion model, we additionally employ content loss and two style losses, i.e., AdaIN style loss and contrastive style loss, aiming to balance the trade-off between style migration and content preservation. Compared with the state-of-the-art models from related fields, our PHDiffusion can stylize the foreground more sufficiently and simultaneously retain finer content. Our code and model are available at https://github.com/bcmi/PHDiffusion-Painterly-Image-Harmonization.
This work focuses on developing methods for approximating the solution operators of a class of parametric partial differential equations via neural operators. Neural operators have several challenges, including the issue of generating appropriate training data, cost-accuracy trade-offs, and nontrivial hyperparameter tuning. The unpredictability of the accuracy of neural operators impacts their applications in downstream problems of inference, optimization, and control. A framework based on the linear variational problem that gives the correction to the prediction furnished by neural operators is considered based on earlier work in JCP 486 (2023) 112104. The operator, called Residual-based Error Corrector Operator or simply Corrector Operator, associated with the corrector problem is analyzed further. Numerical results involving a nonlinear reaction-diffusion model in two dimensions with PCANet-type neural operators show almost two orders of increase in the accuracy of approximations when neural operators are corrected using the correction scheme. Further, topology optimization involving a nonlinear reaction-diffusion model is considered to highlight the limitations of neural operators and the efficacy of the correction scheme. Optimizers with neural operator surrogates are seen to make significant errors (as high as 80 percent). However, the errors are much lower (below 7 percent) when neural operators are corrected.
Well-trained deep neural networks (DNNs) treat all test samples equally during prediction. Adaptive DNN inference with early exiting leverages the observation that some test examples can be easier to predict than others. This paper presents EENet, a novel early-exiting scheduling framework for multi-exit DNN models. Instead of having every sample go through all DNN layers during prediction, EENet learns an early exit scheduler, which can intelligently terminate the inference earlier for certain predictions, which the model has high confidence of early exit. As opposed to previous early-exiting solutions with heuristics-based methods, our EENet framework optimizes an early-exiting policy to maximize model accuracy while satisfying the given per-sample average inference budget. Extensive experiments are conducted on four computer vision datasets (CIFAR-10, CIFAR-100, ImageNet, Cityscapes) and two NLP datasets (SST-2, AgNews). The results demonstrate that the adaptive inference by EENet can outperform the representative existing early exit techniques. We also perform a detailed visualization analysis of the comparison results to interpret the benefits of EENet.
The article reviews significant advances in networked signal and information processing, which have enabled in the last 25 years extending decision making and inference, optimization, control, and learning to the increasingly ubiquitous environments of distributed agents. As these interacting agents cooperate, new collective behaviors emerge from local decisions and actions. Moreover, and significantly, theory and applications show that networked agents, through cooperation and sharing, are able to match the performance of cloud or federated solutions, while offering the potential for improved privacy, increasing resilience, and saving resources.
This paper looks at the task of network topology inference, where the goal is to learn an unknown graph from nodal observations. One of the novelties of the approach put forth is the consideration of prior information about the density of motifs of the unknown graph to enhance the inference of classical Gaussian graphical models. Dealing with the density of motifs directly constitutes a challenging combinatorial task. However, we note that if two graphs have similar motif densities, one can show that the expected value of a polynomial applied to their empirical spectral distributions will be similar. Guided by this, we first assume that we have a reference graph that is related to the sought graph (in the sense of having similar motif densities) and then, we exploit this relation by incorporating a similarity constraint and a regularization term in the network topology inference optimization problem. The (non-)convexity of the optimization problem is discussed and a computational efficient alternating majorization-minimization algorithm is designed. We assess the performance of the proposed method through exhaustive numerical experiments where different constraints are considered and compared against popular baselines algorithms on both synthetic and real-world datasets.
We observe that human poses exhibit strong group-wise structural correlation and spatial coupling between keypoints due to the biological constraints of different body parts. This group-wise structural correlation can be explored to improve the accuracy and robustness of human pose estimation. In this work, we develop a self-constrained prediction-verification network to characterize and learn the structural correlation between keypoints during training. During the inference stage, the feedback information from the verification network allows us to perform further optimization of pose prediction, which significantly improves the performance of human pose estimation. Specifically, we partition the keypoints into groups according to the biological structure of human body. Within each group, the keypoints are further partitioned into two subsets, high-confidence base keypoints and low-confidence terminal keypoints. We develop a self-constrained prediction-verification network to perform forward and backward predictions between these keypoint subsets. One fundamental challenge in pose estimation, as well as in generic prediction tasks, is that there is no mechanism for us to verify if the obtained pose estimation or prediction results are accurate or not, since the ground truth is not available. Once successfully learned, the verification network serves as an accuracy verification module for the forward pose prediction. During the inference stage, it can be used to guide the local optimization of the pose estimation results of low-confidence keypoints with the self-constrained loss on high-confidence keypoints as the objective function. Our extensive experimental results on benchmark MS COCO and CrowdPose datasets demonstrate that the proposed method can significantly improve the pose estimation results.