Voice assistants increasingly use on-device Automatic Speech Recognition (ASR) to ensure speed and privacy. However, due to resource constraints on the device, queries pertaining to complex information domains often require further processing by a search engine. For such applications, we propose a novel Transformer based model capable of rescoring and rewriting, by exploring full context of the N-best hypotheses in parallel. We also propose a new discriminative sequence training objective that can work well for both rescore and rewrite tasks. We show that our Rescore+Rewrite model outperforms the Rescore-only baseline, and achieves up to an average 8.6% relative Word Error Rate (WER) reduction over the ASR system by itself.

Pretraining egocentric vision-language models has become essential to improving downstream egocentric video-text tasks. These egocentric foundation models commonly use the transformer architecture. The memory footprint of these models during pretraining can be substantial. Therefore, we pretrain SViTT-Ego, the first sparse egocentric video-text transformer model integrating edge and node sparsification. We pretrain on the EgoClip dataset and incorporate the egocentric-friendly objective EgoNCE, instead of the frequently used InfoNCE. Most notably, SViTT-Ego obtains a +2.8% gain on EgoMCQ (intra-video) accuracy compared to LAVILA large, with no additional data augmentation techniques other than standard image augmentations, yet pretrainable on memory-limited devices.

When humans create sculptures, we are able to reason about how geometrically we need to alter the clay state to reach our target goal. We are not computing point-wise similarity metrics, or reasoning about low-level positioning of our tools, but instead determining the higher-level changes that need to be made. In this work, we propose LLM-Craft, a novel pipeline that leverages large language models (LLMs) to iteratively reason about and generate deformation-based crafting action sequences. We simplify and couple the state and action representations to further encourage shape-based reasoning. To the best of our knowledge, LLM-Craft is the first system successfully leveraging LLMs for complex deformable object interactions. Through our experiments, we demonstrate that with the LLM-Craft framework, LLMs are able to successfully reason about the deformation behavior of elasto-plastic objects. Furthermore, we find that LLM-Craft is able to successfully create a set of simple letter shapes. Finally, we explore extending the framework to reaching more ambiguous semantic goals, such as "thinner" or "bumpy". For videos please see our website: https://sites.google.com/andrew.cmu.edu/llmcraft.

Compositional learning, mastering the ability to combine basic concepts and construct more intricate ones, is crucial for human cognition, especially in human language comprehension and visual perception. This notion is tightly connected to generalization over unobserved situations. Despite its integral role in intelligence, there is a lack of systematic theoretical and experimental research methodologies, making it difficult to analyze the compositional learning abilities of computational models. In this paper, we survey the literature on compositional learning of AI models and the connections made to cognitive studies. We identify abstract concepts of compositionality in cognitive and linguistic studies and connect these to the computational challenges faced by language and vision models in compositional reasoning. We overview the formal definitions, tasks, evaluation benchmarks, variety of computational models, and theoretical findings. We cover modern studies on large language models to provide a deeper understanding of the cutting-edge compositional capabilities exhibited by state-of-the-art AI models and pinpoint important directions for future research.

We introduce a pipeline that leverages Large Language Models (LLMs) to transform single-turn psychotherapy counseling sessions into multi-turn interactions. While AI-supported online counseling services for individuals with mental disorders exist, they are often constrained by the limited availability of multi-turn training datasets and frequently fail to fully utilize therapists' expertise. Our proposed pipeline effectively addresses these limitations. The pipeline comprises two main steps: 1) Information Extraction and 2) Multi-turn Counseling Generation. Each step is meticulously designed to extract and generate comprehensive multi-turn counseling conversations from the available datasets. Experimental results from both zero-shot and few-shot generation scenarios demonstrate that our approach significantly enhances the ability of LLMs to produce higher quality multi-turn dialogues in the context of mental health counseling. Our pipeline and dataset are publicly available https://github.com/jwkim-chat/A-Data-Augmentation-Pipeline-Leveraging-Large-Language-Models-for-Counseling-Conversations.

Human evaluation serves as the gold standard for assessing the quality of Natural Language Generation (NLG) systems. Nevertheless, the evaluation guideline, as a pivotal element ensuring reliable and reproducible human assessment, has received limited attention.Our investigation revealed that only 29.84% of recent papers involving human evaluation at top conferences release their evaluation guidelines, with vulnerabilities identified in 77.09% of these guidelines. Unreliable evaluation guidelines can yield inaccurate assessment outcomes, potentially impeding the advancement of NLG in the right direction. To address these challenges, we take an initial step towards reliable evaluation guidelines and propose the first human evaluation guideline dataset by collecting annotations of guidelines extracted from existing papers as well as generated via Large Language Models (LLMs). We then introduce a taxonomy of eight vulnerabilities and formulate a principle for composing evaluation guidelines. Furthermore, a method for detecting guideline vulnerabilities has been explored using LLMs, and we offer a set of recommendations to enhance reliability in human evaluation. The annotated human evaluation guideline dataset and code for the vulnerability detection method are publicly available online.

Automatic radiology report generation can significantly benefit the labor-intensive process of report writing by radiologists, especially for 3D radiographs like CT scans, which are crucial for broad clinical diagnostics yet underexplored compared to 2D radiographs. Existing methods often handle 3D volumes either slice-wise or with aggressive downsampling due to current GPU memory limitations, which results in a loss of the inherent 3D nature and critical details. To overcome these issues, we introduce a novel framework that efficiently and effectively generates radiology reports for high-resolution (HR) 3D volumes, based on large language models (LLMs). Specifically, our framework utilizes low-resolution (LR) visual tokens as queries to mine information from HR tokens, preserving detailed HR information while reducing computational costs by only processing HR informed LR visual queries. Further benefiting the field, we curate and release BIMCV-RG, a new dataset with 5,328 HR 3D volumes and paired reports, establishing the first benchmarks for report generation from 3D HR medical images.

With the proliferation of Large Language Model (LLM) based deepfake audio, there is an urgent need for effective detection methods. Previous deepfake audio generation methods typically involve a multi-step generation process, with the final step using a vocoder to predict the waveform from handcrafted features. However, LLM-based audio is directly generated from discrete neural codecs in an end-to-end generation process, skipping the final step of vocoder processing. This poses a significant challenge for current audio deepfake detection (ADD) models based on vocoder artifacts. To effectively detect LLM-based deepfake audio, we focus on the core of the generation process, the conversion from neural codec to waveform. We propose Codecfake dataset, which is generated by seven representative neural codec methods. Experiment results show that codec-trained ADD models exhibit a 41.406% reduction in average equal error rate compared to vocoder-trained ADD models on the Codecfake test set.

Kaplan et al. [2020] (`Kaplan') and Hoffmann et al. [2022] (`Chinchilla') studied the scaling behavior of transformers trained on next-token language prediction. These studies produced different estimates for how the number of parameters ($N$) and training tokens ($D$) should be set to achieve the lowest possible loss for a given compute budget ($C$). Kaplan: $N_\text{optimal} \propto C^{0.73}$, Chinchilla: $N_\text{optimal} \propto C^{0.50}$. This note finds that much of this discrepancy can be attributed to Kaplan counting non-embedding rather than total parameters, combined with their analysis being performed at small scale. Simulating the Chinchilla study under these conditions produces biased scaling coefficients close to Kaplan's. Hence, this note reaffirms Chinchilla's scaling coefficients, by explaining the cause of Kaplan's original overestimation.

With the help of discrete neural audio codecs, large language models (LLM) have increasingly been recognized as a promising methodology for zero-shot Text-to-Speech (TTS) synthesis. However, sampling based decoding strategies bring astonishing diversity to generation, but also pose robustness issues such as typos, omissions and repetition. In addition, the high sampling rate of audio also brings huge computational overhead to the inference process of autoregression. To address these issues, we propose VALL-E R, a robust and efficient zero-shot TTS system, building upon the foundation of VALL-E. Specifically, we introduce a phoneme monotonic alignment strategy to strengthen the connection between phonemes and acoustic sequence, ensuring a more precise alignment by constraining the acoustic tokens to match their associated phonemes. Furthermore, we employ a codec-merging approach to downsample the discrete codes in shallow quantization layer, thereby accelerating the decoding speed while preserving the high quality of speech output. Benefiting from these strategies, VALL-E R obtains controllablity over phonemes and demonstrates its strong robustness by approaching the WER of ground truth. In addition, it requires fewer autoregressive steps, with over 60% time reduction during inference. This research has the potential to be applied to meaningful projects, including the creation of speech for those affected by aphasia. Audio samples will be available at: https://aka.ms/valler.