Speaker-adaptive Text-to-Speech (TTS) synthesis has attracted considerable attention due to its broad range of applications, such as personalized voice assistant services. While several approaches have been proposed, they often exhibit high sensitivity to either the quantity or the quality of target speech samples. To address these limitations, we introduce Stable-TTS, a novel speaker-adaptive TTS framework that leverages a small subset of a high-quality pre-training dataset, referred to as prior samples. Specifically, Stable-TTS achieves prosody consistency by leveraging the high-quality prosody of prior samples, while effectively capturing the timbre of the target speaker. Additionally, it employs a prior-preservation loss during fine-tuning to maintain the synthesis ability for prior samples to prevent overfitting on target samples. Extensive experiments demonstrate the effectiveness of Stable-TTS even under limited amounts of and noisy target speech samples .1

Generating speech from a face image is crucial for developing virtual humans capable of interacting using their unique voices, without relying on pre-recorded human speech. In this paper, we propose Face-StyleSpeech, a zero-shot TextTo-Speech (TTS) synthesis model that generates natural speech conditioned on a face image rather than reference speech. We hypothesize that learning entire prosodic features from a face image poses a significant challenge. To address this, our TTS model incorporates both face and prosody encoders. The prosody encoder is specifically designed to model speech style characteristics that are not fully captured by the face image, allowing the face encoder to focus on extracting speaker-specific features such as timbre. Experimental results demonstrate that Face-StyleSpeech effectively generates more natural speech from a face image than baselines, even for unseen faces. Samples are available on our demo page.1

Safety guard models that detect malicious queries aimed at large language models (LLMs) are essential for ensuring the secure and responsible deployment of LLMs in real-world applications. However, deploying existing safety guard models with billions of parameters alongside LLMs on mobile devices is impractical due to substantial memory requirements and latency. To reduce this cost, we distill a large teacher safety guard model into a smaller one using a labeled dataset of instruction-response pairs with binary harmfulness labels. Due to the limited diversity of harmful instructions in the existing labeled dataset, naively distilled models tend to underperform compared to larger models. To bridge the gap between small and large models, we propose **HarmAug**, a simple yet effective data augmentation method that involves jailbreaking an LLM and prompting it to generate harmful instructions. Given a prompt such as, "Make a single harmful instruction prompt that would elicit offensive content", we add an affirmative prefix (e.g., "I have an idea for a prompt:") to the LLM's response. This encourages the LLM to continue generating the rest of the response, leading to sampling harmful instructions. Another LLM generates a response to the harmful instruction, and the teacher model labels the instruction-response pair. We empirically show that our HarmAug outperforms other relevant baselines. Moreover, a 435-million-parameter safety guard model trained with HarmAug achieves an F1 score comparable to larger models with over 7 billion parameters, and even outperforms them in AUPRC, while operating at less than 25\% of their computational cost. Our [code](https://anonymous.4open.science/r/HarmAug/), [safety guard model](https://huggingface.co/AnonHB/HarmAug_Guard_Model_deberta_v3_large_finetuned), and [synthetic dataset](https://huggingface.co/datasets/AnonHB/HarmAug_generated_dataset) are publicly available.

Large language models (LLMs) excel at complex reasoning tasks but remain computationally expensive, limiting their practical deployment. To address this, recent works have focused on distilling reasoning capabilities into smaller language models (sLMs) using chain-of-thought (CoT) traces from teacher LLMs. However, this approach struggles in scenarios requiring rare factual knowledge or precise computation, where sLMs often hallucinate due to limited capability. In this work, we propose Agent Distillation, a framework for transferring not only reasoning capability but full task-solving behavior from LLM-based agents into sLMs with retrieval and code tools. We improve agent distillation along two complementary axes: (1) we introduce a prompting method called first-thought prefix to enhance the quality of teacher-generated trajectories; and (2) we propose a self-consistent action generation for improving test-time robustness of small agents. We evaluate our method on eight reasoning tasks across factual and mathematical domains, covering both in-domain and out-of-domain generalization. Our results show that sLMs as small as 0.5B, 1.5B, 3B parameters can achieve performance competitive with next-tier larger 1.5B, 3B, 7B models fine-tuned using CoT distillation, demonstrating the potential of agent distillation for building practical, tool-using small agents.

As Large Language Models (LLMs) are increasingly deployed in specialized domains with continuously evolving knowledge, the need for timely and precise knowledge injection has become essential. Fine-tuning with paraphrased data is a common approach to enhance knowledge injection, yet it faces two significant challenges: high computational costs due to repetitive external model usage and limited sample diversity. To this end, we introduce LaPael, a latent-level paraphrasing method that applies input-dependent noise to early LLM layers. This approach enables diverse and semantically consistent augmentations directly within the model. Furthermore, it eliminates the recurring costs of paraphrase generation for each knowledge update. Our extensive experiments on question-answering benchmarks demonstrate that LaPael improves knowledge injection over standard fine-tuning and existing noise-based approaches. Additionally, combining LaPael with data-level paraphrasing further enhances performance.

In cloud environments, GPU-based deep neural network (DNN) inference servers are required to meet the Service Level Objective (SLO) latency for each workload under a specified request rate, while also minimizing GPU resource consumption. However, previous studies have not fully achieved this objective. In this paper, we propose ParvaGPU, a technology that facilitates spatial GPU sharing for large-scale DNN inference in cloud computing. ParvaGPU integrates NVIDIA’s Multi-Instance GPU (MIG) and Multi-Process Service (MPS) technologies to enhance GPU utilization, with the goal of meeting the diverse SLOs of each workload and reducing overall GPU usage. Specifically, ParvaGPU addresses the challenges of minimizing underutilization within allocated GPU space partitions and external fragmentation in combined MIG and MPS environments. We conducted our assessment on multiple A100 GPUs, evaluating 11 diverse DNN workloads with varying SLOs. Our evaluation revealed no SLO violations and a significant reduction in GPU usage compared to state-of-the-art frameworks.

Recent Language Models (LMs) have shown impressive capabilities in generating texts with the knowledge internalized in parameters.Yet, LMs often generate the factually incorrect responses to the given queries, since their knowledge may be inaccurate, incomplete, and outdated.To address this problem, previous works propose to augment LMs with the knowledge retrieved from an external knowledge source.However, such approaches often show suboptimal text generation performance due to two reasons: 1) the model may fail to retrieve the knowledge relevant to the given query, or 2) the model may not faithfully reflect the retrieved knowledge in the generated text.To overcome these, we propose to verify the output and the knowledge of the knowledge-augmented LMs with a separate verifier, which is a small LM that is trained to detect those two types of errors through instruction-finetuning.Then, when the verifier recognizes an error, we can rectify it by either retrieving new knowledge or generating new text.Further, we use an ensemble of the outputs from different instructions with a single verifier to enhance the reliability of the verification processes.We validate the effectiveness of the proposed verification steps on multiple question answering benchmarks, whose results show that the proposed verifier effectively identifies retrieval and generation errors, allowing LMs to provide more factually correct outputs.Our code is available at https://github.com/JinheonBaek/KALMV.

There has been a significant progress in Text-To-Speech (TTS) synthesis technology in recent years, thanks to the advancement in neural generative modeling. However, existing methods on any-speaker adaptive TTS have achieved unsatisfactory performance, due to their suboptimal accuracy in mimicking the target speakers’ styles. In this work, we present Grad-StyleSpeech, which is an any-speaker adaptive TTS framework that is based on a diffusion model that can generate highly natural speech with extremely high similarity to target speakers’ voice, given a few seconds of reference speech. Grad-StyleSpeech significantly outperforms recent speaker-adaptive TTS baselines on English benchmarks. Audio samples are available at https://nardien.github.io/grad-stylespeech-demo.

Large Language Models (LLMs) have shown promising performance in knowledge-intensive reasoning tasks that require a compound understanding of knowledge. However, deployment of the LLMs in real-world applications can be challenging due to their high computational requirements and concerns on data privacy. Previous studies have focused on building task-specific small Language Models (LMs) by fine-tuning them with labeled data or distilling LLMs. However, these approaches are ill-suited for knowledge-intensive reasoning tasks due to the limited capacity of small LMs in memorizing the knowledge required. Motivated by our theoretical analysis on memorization, we propose Knowledge-Augmented Reasoning Distillation (KARD), a novel method that fine-tunes small LMs to generate rationales obtained from LLMs with augmented knowledge retrieved from an external knowledge base. Moreover, we further propose a neural reranker to obtain documents relevant to rationale generation. We empirically show that KARD significantly improves the performance of small T5 and GPT models on the challenging knowledge-intensive reasoning datasets, namely MedQA-USMLE, StrategyQA, and OpenbookQA. Notably, our method makes the 250M T5 models achieve superior performance against the fine-tuned 3B models, having 12 times larger parameters, on both MedQA-USMLE and StrategyQA benchmarks.

Pre-trained language models (PLMs) have achieved remarkable success on various natural language understanding tasks.Simple fine-tuning of PLMs, on the other hand, might be suboptimal for domain-specific tasks because they cannot possibly cover knowledge from all domains.While adaptive pre-training of PLMs can help them obtain domain-specific knowledge, it requires a large training cost.Moreover, adaptive pre-training can harm the PLM's performance on the downstream task by causing catastrophic forgetting of its general knowledge.To overcome such limitations of adaptive pre-training for PLM adaption, we propose a novel domain adaption framework for PLMs coined as Knowledge-Augmented Language model Adaptation (KALA), which modulates the intermediate hidden representations of PLMs with domain knowledge, consisting of entities and their relational facts.We validate the performance of our KALA on question answering and named entity recognition tasks on multiple datasets across various domains.The results show that, despite being computationally efficient, our KALA largely outperforms adaptive pre-training.Code is available at

Seanie Lee, Minki Kang, Juho Lee, Sung Ju Hwang. Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2021.

Graph neural networks have recently achieved remarkable success in representing graph-structured data, with rapid progress in both the node embedding and graph pooling methods. Yet, they mostly focus on capturing information from the nodes considering their connectivity, and not much work has been done in representing the edges, which are essential components of a graph. However, for tasks such as graph reconstruction and generation, as well as graph classification tasks for which the edges are important for discrimination, accurately representing edges of a given graph is crucial to the success of the graph representation learning. To this end, we propose a novel edge representation learning framework based on Dual Hypergraph Transformation (DHT), which transforms the edges of a graph into the nodes of a hypergraph. This dual hypergraph construction allows us to apply message-passing techniques for node representations to edges. After obtaining edge representations from the hypergraphs, we then cluster or drop edges to obtain holistic graph-level edge representations. We validate our edge representation learning method with hypergraphs on diverse graph datasets for graph representation and generation performance, on which our method largely outperforms existing graph representation learning methods. Moreover, our edge representation learning and pooling method also largely outperforms state-of-the-art graph pooling methods on graph classification, not only because of its accurate edge representation learning, but also due to its lossless compression of the nodes and removal of irrelevant edges for effective message-passing.

We propose a method to automatically generate a domain- and task-adaptive maskings of the given text for self-supervised pre-training, such that we can effectively adapt the language model to a particular target task (e.g. question answering). Specifically, we present a novel reinforcement learning-based framework which learns the masking policy, such that using the generated masks for further pre-training of the target language model helps improve task performance on unseen texts. We use off-policy actor-critic with entropy regularization and experience replay for reinforcement learning, and propose a Transformer-based policy network that can consider the relative importance of words in a given text. We validate our Neural Mask Generator (NMG) on several question answering and text classification datasets using BERT and DistilBERT as the language models, on which it outperforms rule-based masking strategies, by automatically learning optimal adaptive maskings.

We consider a novel question answering (QA) task where the machine needs to read from large streaming data (long documents or videos) without knowing when the questions will be given, which is difficult to solve with existing QA methods due to their lack of scalability.To tackle this problem, we propose a novel end-to-end deep network model for reading comprehension, which we refer to as Episodic Memory Reader (EMR) that sequentially reads the input contexts into an external memory, while replacing memories that are less important for answering unseen questions.Specifically, we train an RL agent to replace a memory entry when the memory is full, in order to maximize its QA accuracy at a future timepoint, while encoding the external memory using either the GRU or the Transformer architecture to learn representations that considers relative importance between the memory entries.We validate our model on a synthetic dataset (bAbI) as well as real-world large-scale textual QA (TriviaQA) and video QA (TVQA) datasets, on which it achieves significant improvements over rulebased memory scheduling policies or an RLbased baseline that independently learns the query-specific importance of each memory.