中央研究院 資訊科學研究所

Abstract

Some top-down problem specifications, if executed, may compute sub-problems repeatedly. Instead, we may want a bottom-up algorithm that stores solutions of sub-problems in a table to be reused. How the table can be represented and efficiently maintained, however, can be tricky. We study a special case: computing a function h taking lists as inputs such that hxs is defined in terms of all immediate sublists of xs. Richard Bird studied this problem in 2008 and presented a concise but cryptic algorithm without much explanation. We give this algorithm a proper derivation and discovered a key property that allows it to work. The algorithm builds trees that have certain shapes—the sizes along the left spine is a prefix of a diagonal in Pascal’s triangle. The crucial function we derive transforms one diagonal to the next.

Abstract

In modern Deep Learning, it has been a trend to design larger Deep Neural Networks (DNNs) for the execution of more complex tasks and better accuracy. On the other hand, Convolutional Neural Networks (CNNs) have become the standard method for most of computer vision tasks. However, the memory allocation for the intermediate data in convolution layers can cause severe memory pressure during model training. Many solutions have been proposed to resolve the problem. Besides hardware-dependent solutions, a general methodology rematerialization can reduce GPU memory usage by trading computation for memory efficiently. The idea is to select a set of intermediate results during the forward phase as checkpoints, and only save them in memory to reduce memory usage. The backward phase recomputes the intermediate data from the closest checkpoints in memory as needed. This recomputation increases execution time but saves memory by not storing all intermediate results in memory during the forward phase. In this paper, we will focus on efficiently finding the optimal checkpoint subset to achieve the least peak memory usage during the model training. We first describe the theoretical background of the training of a neural network using mathematical equations. We use these equations to identify all essential data required during both forward and backward phases to compute the gradient of weights of the model. We first identify the checkpoint selection problem and propose a dynamic programming algorithm with time complexity to solve the problem of finding the optimal checkpoint subset. With extensive experiments, we formulate a more accurate description of the problem using our theoretical analysis and revise the objective function based on the tracing, and propose an -time algorithm for finding the optimal checkpoint subset.

Abstract

Federated learning for pathological whole slide image (WSI) classification allows multiple clients to train a global multiple instance learning (MIL) model without sharing their privacy-sensitive WSIs. To accommodate the non-independent and identically distributed (non-i.i.d.) feature shifts, cross-client style transfer has been popularly used but is subject to two fundamental issues: (1) WSIs contain multiple morphological structures due to tissue heterogeneity, and (2) the region of interests (RoIs) is not guaranteed, particularly after augmenting local WSIs data trough style transfer. To address these challenges, we propose HistoFS, a federated learning framework for computational pathology on non-i.i.d. feature shifts in WSI classification. Specifically, we introduce pseudo bag styles that capture multiple style variations within a single WSI. In addition, an authenticity module is introduced to ensure that RoIs are preserved, allowing local models to learn WSIs with diverse styles while maintaining essential RoIs. Extensive experiments validate the superiority of HistoFS over state-of-the-art methods on three clinical datasets.

Abstract

Planning with options -- a sequence of primitive actions -- has been shown effective in reinforcement learning within complex environments. Previous studies have focused on planning with predefined options or learned options through expert demonstration data. Inspired by MuZero, which learns superhuman heuristics without any human knowledge, we propose a novel approach, named OptionZero. OptionZero incorporates an option network into MuZero, providing autonomous discovery of options through self-play games. Furthermore, we modify the dynamics network to provide environment transitions when using options, allowing searching deeper under the same simulation constraints. Empirical experiments conducted in 26 Atari games demonstrate that OptionZero outperforms MuZero, achieving a 131.58% improvement in mean human-normalized score. Our behavior analysis shows that OptionZero not only learns options but also acquires strategic skills tailored to different game characteristics. Our findings show promising directions for discovering and using options in planning. Our code is available at https://rlg.iis.sinica.edu.tw/papers/optionzero.

Abstract

Strength estimation and adjustment are crucial in designing human-AI interactions, particularly in games where AI surpasses human players. This paper introduces a novel strength system, including a strength estimator (SE) and an SE-based Monte Carlo tree search, denoted as SE-MCTS, which predicts strengths from games and offers different playing strengths with human styles. The strength estimator calculates strength scores and predicts ranks from games without direct human interaction. SE-MCTS utilizes the strength scores in a Monte Carlo tree search to adjust playing strength and style. We first conduct experiments in Go, a challenging board game with a wide range of ranks. Our strength estimator significantly achieves over 80% accuracy in predicting ranks by observing 15 games only, whereas the previous method reached 49% accuracy for 100 games. For strength adjustment, SE-MCTS successfully adjusts to designated ranks while achieving a 51.33% accuracy in aligning to human actions, outperforming a previous state-of-the-art, with only 42.56% accuracy. To demonstrate the generality of our strength system, we further apply SE and SE-MCTS to chess and obtain consistent results. These results show a promising approach to strength estimation and adjustment, enhancing human-AI interactions in games. Our code is available at https://rlg.iis.sinica.edu.tw/papers/strength-estimator.

Abstract

As neural network models become gigantic, they increasingly demand more time and memory for training. To meet these demands, advanced parallel computing techniques have become essential. Our research focuses on hybrid parallelism, an extension of pipeline parallelism. Pipeline parallelism splits the neural network into sub-networks distributed across a sequence of processing units, enabling simultaneous processing of different data segments on each device. Hybrid parallelism extends this concept by allocating multiple devices to each sub-network. Our research focuses on optimizing hybrid parallelism by improving how the model is partitioned and how computational devices are assigned. We address these issues by modeling the neural network as a directed acyclic graph of tensor operators, and then demonstrating that optimally partitioning this graph is NP-complete. Then, we propose a two-step approach. The first step is to determine a sequence of nodes. The second step is dynamic programming, which partitions the sequence to maintain balance across the assigned devices. In transforming the graph into a sequence, we explore two methods: one employs topological sorting, while the other clusters non-sequential subgraphs. We apply both methods and select the more effective one based on performance outcomes. We implement our algorithm and conduct experiments. The results show substantial enhancements in both the speed of partitioning and training throughput, with speedups reaching up to 23 in partitioning time and a 1.3-fold increase in training throughput.

Abstract

This work investigates two strategies for zero-shot non-intrusive speech assessment leveraging large language models. First, we explore the audio analysis capabilities of GPT4o. Second, we propose GPT-Whisper, which uses Whisper as an audio-to-text module and evaluates the text’s naturalness via targeted prompt engineering. We evaluate the assessment metrics predicted by GPT-4o and GPT-Whisper, examining their correlation with human-based quality and intelligibility assessments and the character error rate (CER) of automatic speech recognition. Experimental results show that GPT-4o alone is less effective for audio analysis, while GPT-Whisper achieves higher prediction accuracy, has moderate correlation with speech quality and intelligibility, and has higher correlation with CER. Compared to SpeechLMScore, DNSMOS, and VQScore, GPT-Whisper excels in intelligibility metrics, but performs slightly worse than SpeechLMScore in quality estimation. Furthermore, GPTWhisper outperforms supervised non-intrusive models MOS-SSL and MTI-Net in Spearman’s rank correlation for Whisper’s CER. These findings validate GPT-Whisper’s potential for zero-shot speech assessment without requiring additional training data.

Abstract

While pre-trained automatic speech recognition (ASR) systems demonstrate impressive performance on matched domains, their performance often degrades when confronted with channel mismatch stemming from unseen recording environments and conditions. To mitigate this issue, we propose a novel channel-aware data simulation method for robust ASR training. Our method harnesses the synergistic power of channel-extractive techniques and generative adversarial networks (GANs). We first train a channel encoder capable of extracting embeddings from arbitrary audio. On top of this, channel embeddings are extracted using a minimal amount of target-domain data and used to guide a GAN-based speech synthesizer. This synthesizer generates speech that faithfully preserves the phonetic content of the input while mimicking the channel characteristics of the target domain. We evaluate our method on the challenging Hakka Across Taiwan (HAT) and Taiwanese Across Taiwan (TAT) corpora, achieving relative character error rate (CER) reductions of 20.02% and 9.64%, respectively, compared to the baselines. These results highlight the efficacy of our channel-aware data simulation method for bridging the gap between source- and target-domain acoustics.

Abstract

In multichannel speech enhancement, effectively capturing spatial and spectral information across different microphones is crucial for noise reduction. Traditional methods, such as CNN or LSTM, attempt to model the temporal dynamics of fullband and sub-band spectral and spatial features. However, these approaches face limitations in fully modeling complex temporal dependencies, especially in dynamic acoustic environments. To overcome these challenges, we modify the current advanced model McNet by introducing an improved version of Mamba, a state-space model, and further propose MCMamba. MCMamba has been completely reengineered to integrate full-band and narrow-band spatial information with sub-band and full-band spectral features, providing a more comprehensive approach to modeling spatial and spectral information. Our experimental results demonstrate that MCMamba significantly improves the modeling of spatial and spectral features in multichannel speech enhancement, outperforming McNet and achieving very promising performance on the CHiME-3 dataset. Additionally, we find that Mamba performs exceptionally well in modeling spectral information.

Abstract

Hexabromocyclododecane (HBCD) poses significant environmental risks, and identifying HBCD-degrading microbes and their enzymatic mechanisms is challenging due to the complexity of microbial interactions and metabolic pathways. This study aimed to identify critical genes involved in HBCD biodegradation through two approaches: functional annotation of metagenomes and the interpretation of machine learning-based prediction models. Our functional analysis revealed a rich metabolic potential in Chiang Chun soil (CCS) metagenomes, particularly in carbohydrate metabolism. Among the machine learning algorithms tested, random forest models outperformed others, especially when trained on datasets reflecting the degradation patterns of species like Dehalococcoides mccartyi and Pseudomonas aeruginosa. These models highlighted enzymes such as EC 1.8.3.2 (thiol oxidase) and EC 4.1.1.43 (phenylpyruvate decarboxylase) as inhibitors of degradation, while EC 2.7.1.83 (pseudouridine kinase) was linked to enhanced degradation. This dual-methodology approach not only deepens our understanding of microbial functions in HBCD degradation but also provides an unbiased view of the microbial and enzymatic interactions involved, offering a more targeted and effective bioremediation strategy.

Abstract

This paper presents a practical problem in dialogue systems: the capability to adapt to changing user intentions and resolve inconsistencies in conversation histories. It is crucial in scenarios like train ticket booking, where travel plans often change dynamically. Notwithstanding the advancements in NLP and large language models (LLMs), these systems struggle with real-time information updates during conversations. We introduce a specialized dataset to evaluate LLM-based chatbots on such conversational adaptability by asking a broad range of open-domain questions, focusing on scenarios where users modify their requests mid-conversation. Additionally, as LLMs are susceptible to generating superfluous sentences, we propose a novel, Chain-of-Thought-free evaluation framework to distill the user intention from their responses. Through extensive investigations on four LLMs, we observe that these contemporary LLMs are not well-aligned with the latest user intent in long-term conversations; they often fail to capture the nuances of natural conversations in a zero-shot setting. Interestingly, the results demonstrate that GPT-4, widely recognized as having the most advanced reasoning capabilities to date, is bested by GPT-3.5 in this task. This work aims to improve the practicality of LLM-based chatbots, bridging the gap between the current capabilities of dialogue systems and the fluidity of human interactions.

Abstract

Abstract

Abstract

Abstract

Modern AI inference accelerators offer high-performance and power-efficient computations for machine learning models. Most accelerators employ static inference to enhance performance, which requires models to be compiled with predetermined input batch sizes and intermediate tensor shapes. However, static inference can lead to program failures or inefficient execution when processing batched data of varying sizes, a scenario known as dynamic batch inference. This work addresses this challenge by focusing on the emerging multicore AI inference accelerators that offer flexible compute core assignment. We propose to dynamically partition the input batch data into smaller batches, and create multiple model instances to process each partition in parallel. The challenge lies in how to determine the optimal number of model instances, the proper batch size for each handling model, and the assignment of compute cores among the models, to minimize the inference time. To solve the problem, we construct an accurate profiling-based cost model and devise a dynamic programming algorithm to determine the best configuration. Experimental results indicate that our method achieves 3.05× higher throughput on average in multi- person pose estimation benchmarks, compared to the EdgeTPU-like inference strategy.

Abstract

Because a reference signal is often unavailable in real-world scenarios, reference-free speech quality and intelligibility assessment models are important for many speech processing applications. Despite a great number of deep-learning models that have been applied to build non-intrusive speech assessment approaches and achieve promising performance, studies focusing on the hearing impaired (HI) subjects are limited. This paper presents HASA-Net+, a multi-objective non-intrusive hearing-aid speech assessment model, building upon our previous work, HASA-Net. HASA-Net+ improves HASA-Net in several ways: (1) inclusivity for both normal-hearing and HI listeners, (2) integration with pre-trained speech foundation models and fine-tuning techniques, (3) expansion of predictive capabilities to cover speech quality and intelligibility in diverse conditions, including noisy, denoised, reverberant, dereverberated, and vocoded speech, thereby evaluating its robustness, and (4) validation of the generalization capability using an out-of-domain dataset.

Abstract

Background

Primary ovarian insufficiency (POI) is an early decline in ovarian function that leads to ovarian failure. Conventional treatments for POI are inadequate, and treatments based on mesenchymal stem cells (MSCs) have emerged as an option. However, the lack of consideration of the estrogen niche in ovarian tissue significantly reduces the therapeutic efficacy, with an unclear mechanism in the MSCs in POI treatment. Furthermore, the disruption of circadian rhythm associated with POI has not been previously addressed.

Methods

Conditioned medium (CM) and estradiol-conditioned medium (E2-CM) were generated from estrogen receptor positive MSCs (ER+pcMSCs). Chemotherapy-induced POI models were established using C57BL/6 mice (in vivo) and KGN cells (in vitro) treated with cyclophosphamide (CTX) or 4-hydroperoxycyclophosphamide (4-OOH-CP). Gene/protein expressions were detected using RT-qPCR, Western blotting, and immunohistochemistry assays. Locomotor activity was monitored for behavioral circadian rhythmicity. Cytokine arrays and miRNA analysis were conducted to analyze potential factors within CM/E2-CM.

Results

The secretome of ER+pcMSCs (CM and E2-CM) significantly reduced the CTX-induced defects in ovarian folliculogenesis and circadian rhythm. CM/E2-CM also reduced granulosa cell apoptosis and rescued angiogenesis in POI ovarian tissues. E2-CM had a more favorable effect than the CM. Notably, ER+pcMSC secretome restored CTX-induced circadian rhythm defects, including the gene expressions associated with the ovarian circadian clock (e.g., Rora, E4bp4, Rev-erbα, Per2 and Dbp) and locomotor activity. Additionally, the cytokine array analysis revealed a significant increase in cytokines and growth factors associated with immunomodulation and angiogenesis, including angiogenin. Neutralizing the angiogenin in CM/E2-CM significantly reduced its ability to promote HUVEC tube formation in vitro. Exosomal miRNA analysis revealed the miRNAs involved in targeting the genes associated with POI rescue (PTEN and PDCD4), apoptosis (caspase-3, BIM), estrogen synthesis (CYP19A1), ovarian clock regulation (E4BP4, REV-ERBα) and fibrosis (COL1A1).

Conclusion

This study is the first to demonstrate that, in considering the estrogen niche in ovarian tissue, an estrogen-priming ER+pcMSC secretome achieved ovarian regeneration and restored the circadian rhythm in a CTX-induced POI mouse model. The potential factors involved include angiogenin and exosomal miRNAs in the ER+pcMSC secretome. These findings offer insights into potential stem cell therapies for chemotherapy-induced POI and circadian rhythm disruption.

Abstract

Background Alternative splicing is a pivotal mechanism of post-transcriptional modification that contributes to the transcriptome plasticity and proteome diversity in metazoan cells. Although many splicing regulations around the exon/intron regions are known, the relationship between promoter-bound transcription factors and the downstream alternative splicing largely remains unexplored. Results In this study, we present computational approaches to unravel the regulatory relationship between promoter-bound transcription factor binding sites (TFBSs) and the splicing patterns. We curated a fine dataset that includes DNase I hypersensitive site sequencing and transcriptomes across fifteen human tissues from ENCODE. Specifically, we proposed different representations of TF binding context and splicing patterns to examine the associations between the promoter and downstream splicing events. While machine learning models demonstrated potential in predicting splicing patterns based on TFBS occupancies, the limitations in the generalization of predicting the splicing forms of singleton genes across diverse tissues was observed with carefully examination using different cross-validation methods. We further investigated the association between alterations in individual TFBS at promoters and shifts in exon splicing efficiency. Our results demonstrate that the convolutional neural network (CNN) models, trained on TF binding changes in the promoters, can predict the changes in splicing patterns. Furthermore, a systemic in silico substitutions analysis on the CNN models highlighted several potential splicing regulators. Notably, using empirical validation using K562 CTCFL shRNA knock-down data, we showed the significant role of CTCFL in splicing regulation. Conclusion In conclusion, our finding highlights the potential role of promoter-bound TFBSs in influencing the regulation of downstream splicing patterns and provides insights for discovering alternative splicing regulations.

Abstract

Monte-Carlo tree search (MCTS) is an influential sequential decision-making algorithm notably employed in AlphaZero. Despite its success, the primary challenge in AlphaZero training lies in its prolonged time-to-solution due to the high latency imposed by the sequential MCTS process. To address this challenge, this paper proposes and evaluates an inter-decision parallelization strategy called speculative MCTS, a new type of parallelism in AlphaZero which implements speculative execution. This approach allows for the parallel execution of future moves before the current MCTS computations are completed, thus reducing the latency. Additionally, we analyze factors contributing to the overall speedup by studying the synergistic effects of speculation and neural network caching in MCTS. We also provide an analytical model that can be used to evaluate the potential of different speculation strategies before they are implemented and deployed. Our empirical findings indicate that the proposed speculative MCTS can reduce training latency by 5.81x in 9x9 Go games. Moreover, our study shows that speculative execution can enhance the NN cache hit rate by 26% during midgame. Overall, our end-to-end evaluation indicates 1.91x speedup in 19x19 Go training time, compared to the state-of-the-art KataGo program.

Abstract

Cross-domain speech enhancement (SE) is often faced with severe challenges due to the scarcity of noise and background information in an unseen target domain, leading to a mismatch between training and test conditions. This study puts forward a novel data simulation method to address this issue, leveraging noise-extractive techniques and generative adversarial networks (GANs) with only limited target noisy speech data. Notably, our method employs a noise encoder to extract noise embeddings from target-domain data. These embeddings aptly guide the generator to synthesize utterances acoustically fitted to the target domain while authentically preserving the phonetic content of the input clean speech. Furthermore, we introduce the notion of dynamic stochastic perturbation, which can inject controlled perturbations into the noise embeddings during inference, thereby enabling the model to generalize well to unseen noise conditions. Experiments on the VoiceBank-DEMAND benchmark dataset demonstrate that our domain-adaptive SE method outperforms an existing strong baseline based on data simulation.

Abstract

We present the third edition of the VoiceMOS Challenge, a scientific initiative designed to advance research into automatic prediction of human speech ratings. There were three tracks. The first track was on predicting the quality of ``zoomed-in'' high-quality samples from speech synthesis systems. The second track was to predict ratings of samples from singing voice synthesis and voice conversion with a large variety of systems, listeners, and languages. The third track was semi-supervised quality prediction for noisy, clean, and enhanced speech, where a very small amount of labeled training data was provided. Among the eight teams from both academia and industry, we found that many were able to outperform the baseline systems. Successful techniques included retrieval-based methods and the use of non-self-supervised representations like spectrograms and pitch histograms. These results showed that the challenge has advanced the field of subjective speech rating prediction.

Abstract

Background: Increasing evidence suggests an important role of the gut microbiome in the pathogenesis of mental disorders, including depression, along the microbiota-gut-brain axis. The interactions between gut microbe composition and neural circuits in late-life depression (LLD) remain to be elucidated.

Methods: We performed fecal 16S rRNA sequencing and resting-state functional magnetic resonance imaging in a case-control cohort of 32 older adults with LLD, defined as major depressive disorder (MDD), and 16 healthy controls (HCs) to characterize the association of gut microbiota and brain functional connectivity (FC). The Hamilton Depression Rating Scale (HAMD) was used to assess depressive symptoms.

Results: At the genus level, the relative abundances of Enterobacter, Akkermansiaceae, Haemophilus, Burkholderia, and Rothia were significantly higher in depressive patients than in HCs. Reduced FC within mood regulation circuits were mainly found in the frontal cortex (such as the right superior and inferior frontal gyrus, right lateral occipital cortex, left middle frontal gyrus, and left caudate) in the depression patients compared with the HCs. The group-characterized gut microbes in HCs and LLD patients showed opposite correlations with seed-based FC, which may account for the aberrant emotion regulation in depressive patients. The abundance of Enterobacter (dominant genus in LLD) was positively correlated with both HAMD scores and group-characterized FC, while Odoribacter (dominant genus in HC) was negatively correlated with both HAMD scores and group-characterized FC.

Conclusion: Significant correlations were identified between depression-characterized gut microbes and brain FC and depression severity, which may contribute to the pathophysiology of depression development in LLD patients.