Automatic speech quality assessment has raised more attention as an alternative or support to traditional perceptual clinical evaluation. However, most research so far only gains good results on simple tasks such as binary classification, largely due to data scarcity. To deal with this challenge, current works tend to segment patients' audio files into many samples to augment the datasets. Nevertheless, this approach has limitations, as it indirectly relates overall audio scores to individual segments. This paper introduces a novel approach where the system learns at the audio level instead of segments despite data scarcity. This paper proposes to use the pre-trained Wav2Vec2 architecture for both SSL, and ASR as feature extractor in speech assessment. Carried out on the HNC dataset, our ASR-driven approach established a new baseline compared with other approaches, obtaining average MSE = 0.73 and MSE = 1.15 for the prediction of intelligibility and severity scores respectively, using only 95 training samples. It shows that the ASR based Wav2Vec2 model brings the best results and may indicate a strong correlation between ASR and speech quality assessment.

Research on Large Language Models (LLMs) has recently witnessed an increasing interest in extending models' context size to better capture dependencies within long documents. While benchmarks have been proposed to assess long-range abilities, existing efforts primarily considered generic tasks that are not necessarily aligned with real-world applications. In contrast, our work proposes a new benchmark for long-context LLMs focused on a practical meeting assistant scenario. In this scenario, the long contexts consist of transcripts obtained by automatic speech recognition, presenting unique challenges for LLMs due to the inherent noisiness and oral nature of such data. Our benchmark, named ELITR-Bench, augments the existing ELITR corpus' transcripts with 271 manually crafted questions and their ground-truth answers. Our experiments with recent long-context LLMs on ELITR-Bench highlight a gap between open-source and proprietary models, especially when questions are asked sequentially within a conversation. We also provide a thorough analysis of our GPT-4-based evaluation method, encompassing insights from a crowdsourcing study. Our findings suggest that while GPT-4's evaluation scores are correlated with human judges', its ability to differentiate among more than three score levels may be limited.

Recent developments in hardware and information technology have enabled the emergence of billions of connected, intelligent devices around the world exchanging information with minimal human involvement. This paradigm, known as the Internet of Things (IoT), is progressing quickly, with an estimated 27 billion devices by 2025 (almost four devices per person) [1, 2]. These smart devices help improve our quality of life, with wearables to monitor health, vehicles that interact with traffic centers and other vehicles to ensure safety, and various home appliances offering comfort. This increase in the number of IoT devices and successful IoT services has generated tremendous data. The International Data Corporation report estimates that by 2025 this data will grow from 4 to 140 zettabytes [3].

The capacity and effectiveness of pre-trained multilingual models (MLMs) for zero-shot cross-lingual transfer is well established. However, phenomena of positive or negative transfer, and the effect of language choice still need to be fully understood, especially in the complex setting of massively multilingual LMs. We propose an efficient method to study transfer language influence in zero-shot performance on another target language. Unlike previous work, our approach disentangles downstream tasks from language, using dedicated adapter units. Our findings suggest that some languages do not largely affect others, while some languages, especially ones unseen during pre-training, can be extremely beneficial or detrimental for different target languages. We find that no transfer language is beneficial for all target languages. We do, curiously, observe languages previously unseen by MLMs consistently benefit from Figure 1: Our approach uses efficient few-step continued transfer from almost any language. We additionally tuning (left) and adapter modules (right) to disentangle use our modular approach to quantify the effect of task and language to quantify the effect negative interference efficiently and catagorize of a transfer language for a given task and model.

Conceptual reasoning, the ability to reason in abstract and high-level perspectives, is key to generalization in human cognition. However, limited study has been done on large language models' capability to perform conceptual reasoning. In this work, we bridge this gap and propose a novel conceptualization framework that forces models to perform conceptual reasoning on abstract questions and generate solutions in a verifiable symbolic space. Using this framework as an analytical tool, we show that existing large language models fall short on conceptual reasoning, dropping 9% to 28% on various benchmarks compared to direct inference methods. We then discuss how models can improve since high-level abstract reasoning is key to unbiased and generalizable decision-making. We propose two techniques to add trustworthy induction signals by generating familiar questions with similar underlying reasoning paths and asking models to perform self-refinement. Experiments show that our proposed techniques improve models' conceptual reasoning performance by 8% to 11%, achieving a more robust reasoning system that relies less on inductive biases.

Counter narratives - informed responses to hate speech contexts designed to refute hateful claims and de-escalate encounters - have emerged as an effective hate speech intervention strategy. While previous work has proposed automatic counter narrative generation methods to aid manual interventions, the evaluation of these approaches remains underdeveloped. Previous automatic metrics for counter narrative evaluation lack alignment with human judgment as they rely on superficial reference comparisons instead of incorporating key aspects of counter narrative quality as evaluation criteria. To address prior evaluation limitations, we propose a novel evaluation framework prompting LLMs to provide scores and feedback for generated counter narrative candidates using 5 defined aspects derived from guidelines from counter narrative specialized NGOs. We found that LLM evaluators achieve strong alignment to human-annotated scores and feedback and outperform alternative metrics, indicating their potential as multi-aspect, reference-free and interpretable evaluators for counter narrative evaluation.

Conventional processes for analyzing datasets and extracting meaningful information are often time-consuming and laborious. Previous work has identified manual, repetitive coding and data collection as major obstacles that hinder data scientists from undertaking more nuanced labor and high-level projects. To combat this, we evaluated OpenAI's GPT-3.5 as a "Language Data Scientist" (LDS) that can extrapolate key findings, including correlations and basic information, from a given dataset. The model was tested on a diverse set of benchmark datasets to evaluate its performance across multiple standards, including data science code-generation based tasks involving libraries such as NumPy, Pandas, Scikit-Learn, and TensorFlow, and was broadly successful in correctly answering a given data science query related to the benchmark dataset. The LDS used various novel prompt engineering techniques to effectively answer a given question, including Chain-of-Thought reinforcement and SayCan prompt engineering. Our findings demonstrate great potential for leveraging Large Language Models for low-level, zero-shot data analysis.

Monocular 3D detectors achieve remarkable performance on cars and smaller objects. However, their performance drops on larger objects, leading to fatal accidents. Some attribute the failures to training data scarcity or their receptive field requirements of large objects. In this paper, we highlight this understudied problem of generalization to large objects. We find that modern frontal detectors struggle to generalize to large objects even on nearly balanced datasets. We argue that the cause of failure is the sensitivity of depth regression losses to noise of larger objects. To bridge this gap, we comprehensively investigate regression and dice losses, examining their robustness under varying error levels and object sizes. We mathematically prove that the dice loss leads to superior noise-robustness and model convergence for large objects compared to regression losses for a simplified case. Leveraging our theoretical insights, we propose SeaBird (Segmentation in Bird's View) as the first step towards generalizing to large objects. SeaBird effectively integrates BEV segmentation on foreground objects for 3D detection, with the segmentation head trained with the dice loss. SeaBird achieves SoTA results on the KITTI-360 leaderboard and improves existing detectors on the nuScenes leaderboard, particularly for large objects. Code and models at https://github.com/abhi1kumar/SeaBird

Scientific articles are long text documents organized into sections, each describing aspects of the research. Analyzing scientific production has become progressively challenging due to the increase in the number of available articles. Within this scenario, our approach consisted of fine-tuning transformer language models to generate sentence-level embeddings from scientific articles, considering the following labels: background, objective, methods, results, and conclusion. We trained our models on three datasets with contrastive learning. Two datasets are from the article's abstracts in the computer science and medical domains. Also, we introduce PMC-Sents-FULL, a novel dataset of sentences extracted from the full texts of medical articles. We compare the fine-tuned and baseline models in clustering and classification tasks to evaluate our approach. On average, clustering agreement measures values were five times higher. For the classification measures, in the best-case scenario, we had an average improvement in F1-micro of 30.73\%. Results show that fine-tuning sentence transformers with contrastive learning and using the generated embeddings in downstream tasks is a feasible approach to sentence classification in scientific articles. Our experiment codes are available on GitHub.

The widespread application of machine learning algorithms is a matter of increasing concern for the data privacy research community, and many have sought to develop privacy-preserving techniques for it. Among existing approaches, the homomorphic evaluation of ML algorithms stands out by performing operations directly over encrypted data, enabling strong guarantees of confidentiality. The homomorphic evaluation of inference algorithms is practical even for relatively deep Convolution Neural Networks (CNNs). However, training is still a major challenge, with current solutions often resorting to lightweight algorithms that can be unfit for solving more complex problems, such as image recognition. This work introduces the homomorphic evaluation of Wilkie, Stonham, and Aleksander's Recognition Device (WiSARD) and subsequent Weightless Neural Networks (WNNs) for training and inference on encrypted data. Compared to CNNs, WNNs offer better performance with a relatively small accuracy drop. We develop a complete framework for it, including several building blocks that can be of independent interest. Our framework achieves 91.7% accuracy on the MNIST dataset after only 3.5 minutes of encrypted training (multi-threaded), going up to 93.8% in 3.5 hours. For the HAM10000 dataset, we achieve 67.9% accuracy in just 1.5 minutes, going up to 69.9% after 1 hour. Compared to the state of the art on the HE evaluation of CNN training, Glyph (Lou et al., NeurIPS 2020), these results represent a speedup of up to 1200 times with an accuracy loss of at most 5.4%. For HAM10000, we even achieved a 0.65% accuracy improvement while being 60 times faster than Glyph. We also provide solutions for small-scale encrypted training. In a single thread on a desktop machine using less than 200MB of memory, we train over 1000 MNIST images in 12 minutes or over the entire Wisconsin Breast Cancer dataset in just 11 seconds.