Discriminativeness is a desirable feature of image captions: captions should describe the characteristic details of input images. However, recent high-performing captioning models, which are trained with reinforcement learning (RL), tend to generate overly generic captions despite their high performance in various other criteria. First, we investigate the cause of the unexpectedly low discriminativeness and show that RL has a deeply rooted side effect of limiting the output words to high-frequency words. The limited vocabulary is a severe bottleneck for discriminativeness as it is difficult for a model to describe the details beyond its vocabulary. Then, based on this identification of the bottleneck, we drastically recast discriminative image captioning as a much simpler task of encouraging low-frequency word generation. Hinted by long-tail classification and debiasing methods, we propose methods that easily switch off-the-shelf RL models to discriminativeness-aware models with only a single-epoch fine-tuning on the part of the parameters. Extensive experiments demonstrate that our methods significantly enhance the discriminativeness of off-the-shelf RL models and even outperform previous discriminativeness-aware methods with much smaller computational costs. Detailed analysis and human evaluation also verify that our methods boost the discriminativeness without sacrificing the overall quality of captions.

多语言神经机器翻译可以在训练过程中翻译看不见的语言对，即零弹性翻译。但是，零拍的翻译总是不稳定的。尽管先前的作品将不稳定归因于中心语言的统治，例如英语，我们以非中心语言的严格依赖性来补充这种观点。在这项工作中，我们通过适应非中心语言并将共享信息和特定于语言的信息组合来抵消零拍的不稳定性，从而提出了一种简单，轻巧但有效的特定语言建模方法。在IWSLT17，Europarl，TED Talks和Opus-100数据集上进行变压器的实验表明，我们的方法不仅在中心数据条件下的性能优于强基础，而且还可以轻松拟合非中心数据条件。通过进一步研究层归因，我们表明我们所提出的方法可以将耦合表示形式朝正确的方向解散。

对于知识图形完成，存在两种主要类型的预测模型：基于图形嵌入的一个，以及基于关系路径规则诱导。它们具有不同的优缺点。为了利用这两种类型，最近提出了混合模型。其中一个混合模型，uniker，交替通过关系路径规则增强培训数据并列进嵌入模型。尽管其预测准确性很高，但它不充分利用关系路径规则，因为它忽略了低置信度规则，以保持增强数据的质量。为了缓解此限制，我们通过关系路径规则和基于置信性的增强数据提出转换数据增强。结果和分析表明，我们所提出的方法通过增强包括与它们类似的真实答案或实体的数据来有效提高嵌入模型的性能。

Hierarchical Reinforcement Learning (HRL) algorithms have been demonstrated to perform well on high-dimensional decision making and robotic control tasks. However, because they solely optimize for rewards, the agent tends to search the same space redundantly. This problem reduces the speed of learning and achieved reward. In this work, we present an Off-Policy HRL algorithm that maximizes entropy for efficient exploration. The algorithm learns a temporally abstracted low-level policy and is able to explore broadly through the addition of entropy to the high-level. The novelty of this work is the theoretical motivation of adding entropy to the RL objective in the HRL setting. We empirically show that the entropy can be added to both levels if the Kullback-Leibler (KL) divergence between consecutive updates of the low-level policy is sufficiently small. We performed an ablative study to analyze the effects of entropy on hierarchy, in which adding entropy to high-level emerged as the most desirable configuration. Furthermore, a higher temperature in the low-level leads to Q-value overestimation and increases the stochasticity of the environment that the high-level operates on, making learning more challenging. Our method, SHIRO, surpasses state-of-the-art performance on a range of simulated robotic control benchmark tasks and requires minimal tuning.

The network architecture of end-to-end (E2E) automatic speech recognition (ASR) can be classified into several models, including connectionist temporal classification (CTC), recurrent neural network transducer (RNN-T), attention mechanism, and non-autoregressive mask-predict models. Since each of these network architectures has pros and cons, a typical use case is to switch these separate models depending on the application requirement, resulting in the increased overhead of maintaining all models. Several methods for integrating two of these complementary models to mitigate the overhead issue have been proposed; however, if we integrate more models, we will further benefit from these complementary models and realize broader applications with a single system. This paper proposes four-decoder joint modeling (4D) of CTC, attention, RNN-T, and mask-predict, which has the following three advantages: 1) The four decoders are jointly trained so that they can be easily switched depending on the application scenarios. 2) Joint training may bring model regularization and improve the model robustness thanks to their complementary properties. 3) Novel one-pass joint decoding methods using CTC, attention, and RNN-T further improves the performance. The experimental results showed that the proposed model consistently reduced the WER.

Spoken language understanding (SLU) tasks have been studied for many decades in the speech research community, but have not received as much attention as lower-level tasks like speech and speaker recognition. In particular, there are not nearly as many SLU task benchmarks, and many of the existing ones use data that is not freely available to all researchers. Recent work has begun to introduce such benchmark datasets for several tasks. In this work, we introduce several new annotated SLU benchmark tasks based on freely available speech data, which complement existing benchmarks and address gaps in the SLU evaluation landscape. We contribute four tasks: question answering and summarization involve inference over longer speech sequences; named entity localization addresses the speech-specific task of locating the targeted content in the signal; dialog act classification identifies the function of a given speech utterance. We follow the blueprint of the Spoken Language Understanding Evaluation (SLUE) benchmark suite. In order to facilitate the development of SLU models that leverage the success of pre-trained speech representations, we will be publishing for each task (i) annotations for a relatively small fine-tuning set, (ii) annotated development and test sets, and (iii) baseline models for easy reproducibility and comparisons. In this work, we present the details of data collection and annotation and the performance of the baseline models. We also perform sensitivity analysis of pipeline models' performance (speech recognizer + text model) to the speech recognition accuracy, using more than 20 state-of-the-art speech recognition models.

Self-supervised pre-trained transformers have improved the state of the art on a variety of speech tasks. Due to the quadratic time and space complexity of self-attention, they usually operate at the level of relatively short (e.g., utterance) segments. In this paper, we study the use of context, i.e., surrounding segments, during fine-tuning and propose a new approach called context-aware fine-tuning. We attach a context module on top of the last layer of a pre-trained model to encode the whole segment into a context embedding vector which is then used as an additional feature for the final prediction. During the fine-tuning stage, we introduce an auxiliary loss that encourages this context embedding vector to be similar to context vectors of surrounding segments. This allows the model to make predictions without access to these surrounding segments at inference time and requires only a tiny overhead compared to standard fine-tuned models. We evaluate the proposed approach using the SLUE and Librilight benchmarks for several downstream tasks: Automatic speech recognition (ASR), named entity recognition (NER), and sentiment analysis (SA). The results show that context-aware fine-tuning not only outperforms a standard fine-tuning baseline but also rivals a strong context injection baseline that uses neighboring speech segments during inference.

Direct speech-to-speech translation (S2ST), in which all components can be optimized jointly, is advantageous over cascaded approaches to achieve fast inference with a simplified pipeline. We present a novel two-pass direct S2ST architecture, {\textit UnitY}, which first generates textual representations and predicts discrete acoustic units subsequently. We enhance the model performance by subword prediction in the first-pass decoder, advanced two-pass decoder architecture design and search strategy, and better training regularization. To leverage large amounts of unlabeled text data, we pre-train the first-pass text decoder based on the self-supervised denoising auto-encoding task. Experimental evaluations on benchmark datasets at various data scales demonstrate that UnitY outperforms a single-pass speech-to-unit translation model by 2.5-4.2 ASR-BLEU with 2.83x decoding speed-up. We show that the proposed methods boost the performance even when predicting spectrogram in the second pass. However, predicting discrete units achieves 2.51x decoding speed-up compared to that case.

Hyperparameter optimization (HPO) is essential for the better performance of deep learning, and practitioners often need to consider the trade-off between multiple metrics, such as error rate, latency, memory requirements, robustness, and algorithmic fairness. Due to this demand and the heavy computation of deep learning, the acceleration of multi-objective (MO) optimization becomes ever more important. Although meta-learning has been extensively studied to speedup HPO, existing methods are not applicable to the MO tree-structured parzen estimator (MO-TPE), a simple yet powerful MO-HPO algorithm. In this paper, we extend TPE's acquisition function to the meta-learning setting, using a task similarity defined by the overlap in promising domains of each task. In a comprehensive set of experiments, we demonstrate that our method accelerates MO-TPE on tabular HPO benchmarks and yields state-of-the-art performance. Our method was also validated externally by winning the AutoML 2022 competition on "Multiobjective Hyperparameter Optimization for Transformers".

While human evaluation is the most reliable metric for evaluating speech generation systems, it is generally costly and time-consuming. Previous studies on automatic speech quality assessment address the problem by predicting human evaluation scores with machine learning models. However, they rely on supervised learning and thus suffer from high annotation costs and domain-shift problems. We propose SpeechLMScore, an unsupervised metric to evaluate generated speech using a speech-language model. SpeechLMScore computes the average log-probability of a speech signal by mapping it into discrete tokens and measures the average probability of generating the sequence of tokens. Therefore, it does not require human annotation and is a highly scalable framework. Evaluation results demonstrate that the proposed metric shows a promising correlation with human evaluation scores on different speech generation tasks including voice conversion, text-to-speech, and speech enhancement.