分享

Stochastic sampling strategies such as top-k and top-p have been widely used in dialogue generation task. However, as an open-domain chatting system, there will be two different conversation scenarios, i.e. chit-chat and knowledge-based question answering. In the former situation, responses diversity is essential due to the one-to-many nature in dialogue. The latter, on the other hand, requires less randomness given that stochastic decoding strategy entails the risk of generating incorrect information. As a result, an adaptive and flexible decoding strategy is needed to cope with these two scenarios simultaneously. To this end, we propose the dynamic decoding strategy (DDS), which can adjust the decoding space w.r.t. different contexts. In DDS, both sequence-level and token-level adaptive search can be achieved to adjust the decoding process in a unified framework. Besides, our adaptive algorithm can not only be used during model inference, but it can also be applied during the model training stage to further enhance the performance. Comprehensive experiments indicate that the proposed decoding strategy can consistently improve the performance of pre-trained dialogue models when coupled with four well-used stochastic decoding algorithms.

Generative artificial intelligence (GenAI) and large language models (LLMs) have simultaneously opened new avenues for enhancing human learning and increased the prevalence of poor-quality information in student response - termed Botpoop. This study introduces Professor Leodar, a custom-built, Singlish-speaking Retrieval Augmented Generation (RAG) chatbot designed to enhance educational while reducing Botpoop. Deployed at Nanyang Technological University, Singapore, Professor Leodar offers a glimpse into the future of AI-assisted learning, offering personalized guidance, 24/7 availability, and contextually relevant information. Through a mixed-methods approach, we examine the impact of Professor Leodar on learning, engagement, and exam preparedness, with 97.1% of participants reporting positive experiences. These findings help define possible roles of AI in education and highlight the potential of custom GenAI chatbots. Our combination of chatbot development, in-class deployment and outcomes study offers a benchmark for GenAI educational tools and is a stepping stone for redefining the interplay between AI and human learning.

Greybox fuzzing has achieved success in revealing bugs and vulnerabilities in programs. However, randomized mutation strategies have limited the fuzzer's performance on structured data. Specialized fuzzers can handle complex structured data, but require additional efforts in grammar and suffer from low throughput. In this paper, we explore the potential of utilizing the Large Language Model to enhance greybox fuzzing for structured data. We utilize the pre-trained knowledge of LLM about data conversion and format to generate new valid inputs. We further fine-tuned it with paired mutation seeds to learn structured format and mutation strategies effectively. Our LLM-based fuzzer, LLAMAFUZZ, integrates the power of LLM to understand and mutate structured data to fuzzing. We conduct experiments on the standard bug-based benchmark Magma and a wide variety of real-world programs. LLAMAFUZZ outperforms our top competitor by 41 bugs on average. We also identified 47 unique bugs across all trials. Moreover, LLAMAFUZZ demonstrated consistent performance on both bug trigger and bug reached. Compared to AFL++, LLAMAFUZZ achieved 27.19% more branches in real-world program sets on average. We also demonstrate a case study to explain how LLMs enhance the fuzzing process in terms of code coverage.

Frontier Large Language Models (LLMs) can be socially discriminatory or sensitive to spurious features of their inputs. Because only well-resourced corporations can train frontier LLMs, we need robust test-time strategies to control such biases. Existing solutions, which instruct the LLM to be fair or robust, rely on the model's implicit understanding of bias. Causality provides a rich formalism through which we can be explicit about our debiasing requirements. Yet, as we show, a naive application of the standard causal debiasing strategy, counterfactual data augmentation, fails under standard assumptions to debias predictions at an individual level at test time. To address this, we develop a stratified notion of debiasing called stratified invariance, which can capture a range of debiasing requirements from population level to individual level through an additional measurement that stratifies the predictions. We present a complete observational test for stratified invariance. Finally, we introduce a data augmentation strategy that guarantees stratified invariance at test time under suitable assumptions, together with a prompting strategy that encourages stratified invariance in LLMs. We show that our prompting strategy, unlike implicit instructions, consistently reduces the bias of frontier LLMs across a suite of synthetic and real-world benchmarks without requiring additional data, finetuning or pre-training.

Speech enhancement significantly improves the clarity and intelligibility of speech in noisy environments, improving communication and listening experiences. In this paper, we introduce a novel pretraining feature-guided diffusion model tailored for efficient speech enhancement, addressing the limitations of existing discriminative and generative models. By integrating spectral features into a variational autoencoder (VAE) and leveraging pre-trained features for guidance during the reverse process, coupled with the utilization of the deterministic discrete integration method (DDIM) to streamline sampling steps, our model improves efficiency and speech enhancement quality. Demonstrating state-of-the-art results on two public datasets with different SNRs, our model outshines other baselines in efficiency and robustness. The proposed method not only optimizes performance but also enhances practical deployment capabilities, without increasing computational demands.

This paper presents a new approach to the problem of correcting speech recognition errors by means of post-editing. It consists of using a neural sequence tagger that learns how to correct an ASR (Automatic Speech Recognition) hypothesis word by word and a corrector module that applies corrections returned by the tagger. The proposed solution is applicable to any ASR system, regardless of its architecture, and provides high-precision control over errors being corrected. This is especially crucial in production environments, where avoiding the introduction of new mistakes by the error correction model may be more important than the net gain in overall results. The results show that the performance of the proposed error correction models is comparable with previous approaches while requiring much smaller resources to train, which makes it suitable for industrial applications, where both inference latency and training times are critical factors that limit the use of other techniques.

Sharpness-Aware Minimization (SAM) has emerged as a promising approach for effectively reducing the generalization error. However, SAM incurs twice the computational cost compared to base optimizer (e.g., SGD). We propose Asymptotic Unbiased Sampling with respect to iterations to accelerate SAM (AUSAM), which maintains the model's generalization capacity while significantly enhancing computational efficiency. Concretely, we probabilistically sample a subset of data points beneficial for SAM optimization based on a theoretically guaranteed criterion, i.e., the Gradient Norm of each Sample (GNS). We further approximate the GNS by the difference in loss values before and after perturbation in SAM. As a plug-and-play, architecture-agnostic method, our approach consistently accelerates SAM across a range of tasks and networks, i.e., classification, human pose estimation and network quantization. On CIFAR10/100 and Tiny-ImageNet, AUSAM achieves results comparable to SAM while providing a speedup of over 70%. Compared to recent dynamic data pruning methods, AUSAM is better suited for SAM and excels in maintaining performance. Additionally, AUSAM accelerates optimization in human pose estimation and model quantization without sacrificing performance, demonstrating its broad practicality.

We present a principled approach to provide LLM-based evaluation with a rigorous guarantee of human agreement. We first propose that a reliable evaluation method should not uncritically rely on model preferences for pairwise evaluation, but rather assess the confidence of judge models and selectively decide when to trust its judgement. We then show that under this selective evaluation framework, human agreement can be provably guaranteed -- such that the model evaluation aligns with that of humans to a user-specified agreement level. As part of our framework, we also introduce Simulated Annotators, a novel confidence estimation method that significantly improves judge calibration and thus enables high coverage of evaluated instances. Finally, we propose Cascaded Selective Evaluation, where we use cheaper models as initial judges and escalate to stronger models only when necessary -- again, while still providing a provable guarantee of human agreement. Experimental results show that Cascaded Selective Evaluation guarantees strong alignment with humans, far beyond what LLM judges could achieve without selective evaluation. For example, on a subset of Chatbot Arena where GPT-4 almost never achieves 80% human agreement, our method, even while employing substantially cost-effective models such as Mistral-7B, guarantees over 80% human agreement with almost 80% test coverage.

Recent advancements in textless speech-to-speech translation systems have been driven by the adoption of self-supervised learning techniques. Although most state-of-the-art systems adopt a similar architecture to transform source language speech into sequences of discrete representations in the target language, the criteria for selecting these target speech units remains an open question. This work explores the selection process through a study of downstream tasks such as automatic speech recognition, speech synthesis, speaker recognition, and emotion recognition. Interestingly, our findings reveal a discrepancy in the optimization of discrete speech units: units that perform well in resynthesis performance do not necessarily correlate with those that enhance translation efficacy. This discrepancy underscores the nuanced complexity of target feature selection and its impact on the overall performance of speech-to-speech translation systems.

Large Language Model (LLMs) such as ChatGPT that exhibit generative AI capabilities are facing accelerated adoption and innovation. The increased presence of Generative AI (GAI) inevitably raises concerns about the risks and safety associated with these models. This article provides an up-to-date survey of recent trends in AI safety research of GAI-LLMs from a computer scientist's perspective: specific and technical. In this survey, we explore the background and motivation for the identified harms and risks in the context of LLMs being generative language models; our survey differentiates by emphasising the need for unified theories of the distinct safety challenges in the research development and applications of LLMs. We start our discussion with a concise introduction to the workings of LLMs, supported by relevant literature. Then we discuss earlier research that has pointed out the fundamental constraints of generative models, or lack of understanding thereof (e.g., performance and safety trade-offs as LLMs scale in number of parameters). We provide a sufficient coverage of LLM alignment -- delving into various approaches, contending methods and present challenges associated with aligning LLMs with human preferences. By highlighting the gaps in the literature and possible implementation oversights, our aim is to create a comprehensive analysis that provides insights for addressing AI safety in LLMs and encourages the development of aligned and secure models. We conclude our survey by discussing future directions of LLMs for AI safety, offering insights into ongoing research in this critical area.

Traditional speech enhancement methods often oversimplify the task of restoration by focusing on a single type of distortion. Generative models that handle multiple distortions frequently struggle with phone reconstruction and high-frequency harmonics, leading to breathing and gasping artifacts that reduce the intelligibility of reconstructed speech. These models are also computationally demanding, and many solutions are restricted to producing outputs in the wide-band frequency range, which limits their suitability for professional applications. To address these challenges, we propose Hi-ResLDM, a novel generative model based on latent diffusion designed to remove multiple distortions and restore speech recordings to studio quality, sampled at 48kHz. We benchmark Hi-ResLDM against state-of-the-art methods that leverage GAN and Conditional Flow Matching (CFM) components, demonstrating superior performance in regenerating high-frequency-band details. Hi-ResLDM not only excels in non-instrusive metrics but is also consistently preferred in human evaluation and performs competitively on intrusive evaluations, making it ideal for high-resolution speech restoration.

This research dissects financial equity research reports (ERRs) by mapping their content into categories. There is insufficient empirical analysis of the questions answered in ERRs. In particular, it is not understood how frequently certain information appears, what information is considered essential, and what information requires human judgment to distill into an ERR. The study analyzes 72 ERRs sentence-by-sentence, classifying their 4940 sentences into 169 unique question archetypes. We did not predefine the questions but derived them solely from the statements in the ERRs. This approach provides an unbiased view of the content of the observed ERRs. Subsequently, we used public corporate reports to classify the questions' potential for automation. Answers were labeled "text-extractable" if the answers to the question were accessible in corporate reports. 78.7% of the questions in ERRs can be automated. Those automatable question consist of 48.2% text-extractable (suited to processing by large language models, LLMs) and 30.5% database-extractable questions. Only 21.3% of questions require human judgment to answer. We empirically validate using Llama-3-70B and GPT-4-turbo-2024-04-09 that recent advances in language generation and information extraction enable the automation of approximately 80% of the statements in ERRs. Surprisingly, the models complement each other's strengths and weaknesses well. The research confirms that the current writing process of ERRs can likely benefit from additional automation, improving quality and efficiency. The research thus allows us to quantify the potential impacts of introducing large language models in the ERR writing process. The full question list, including the archetypes and their frequency, will be made available online after peer review.