Measuring memorization through probabilistic discoverable extraction

Large language models (LLMs) are susceptible to memorizing training data, raising concerns due to the potential extraction of sensitive information. Current methods to measure memorization rates of LLMs, primarily discoverable extraction (Carlini et al., 2022), rely on single-sequence greedy sampling, potentially underestimating the true extent of memorization. This paper introduces a probabilistic relaxation of discoverable extraction that quantifies the probability of extracting a target sequence within a set of generated samples, considering various sampling schemes and multiple attempts. This approach addresses the limitations of reporting memorization rates through discoverable extraction by accounting for the probabilistic nature of LLMs and user interaction patterns. Our experiments demonstrate that this probabilistic measure can reveal cases of higher memorization rates compared to rates found through discoverable extraction. We further investigate the impact of different sampling schemes on extractability, providing a more comprehensive and realistic assessment of LLM memorization and its associated risks. Our contributions include a new probabilistic memorization definition, empirical evidence of its effectiveness, and a thorough evaluation across different models, sizes, sampling schemes, and training data repetitions.

Conformal Prediction for Multimodal Regression

This paper introduces multimodal conformal regression. Traditionally confined to scenarios with solely numerical input features, conformal prediction is now extended to multimodal contexts through our methodology, which harnesses internal features from complex neural network architectures processing images and unstructured text. Our findings highlight the potential for internal neural network features, extracted from convergence points where multimodal information is combined, to be used by conformal prediction to construct prediction intervals (PIs). This capability paves new paths for deploying conformal prediction in domains abundant with multimodal data, enabling a broader range of problems to benefit from guaranteed distribution-free uncertainty quantification.

Assessing Alcohol Use Disorder: Insights from Lifestyle, Background, and Family History with Machine Learning Techniques

This study explored how lifestyle, personal background, and family history contribute to the risk of developing Alcohol Use Disorder (AUD). Survey data from the All of Us Program was utilized to extract information on AUD status, lifestyle, personal background, and family history for 6,016 participants. Key determinants of AUD were identified using decision trees including annual income, recreational drug use, length of residence, sex/gender, marital status, education level, and family history of AUD. Data visualization and Chi-Square Tests of Independence were then used to assess associations between identified factors and AUD. Afterwards, machine learning techniques including decision trees, random forests, and Naive Bayes were applied to predict an individual's likelihood of developing AUD. Random forests were found to achieve the highest accuracy (82%), compared to Decision Trees and Naive Bayes. Findings from this study can offer insights that help parents, healthcare professionals, and educators develop strategies to reduce AUD risk, enabling early intervention and targeted prevention efforts.

MMAU: A Massive Multi-Task Audio Understanding and Reasoning Benchmark

The ability to comprehend audio--which includes speech, non-speech sounds, and music--is crucial for AI agents to interact effectively with the world. We present MMAU, a novel benchmark designed to evaluate multimodal audio understanding models on tasks requiring expert-level knowledge and complex reasoning. MMAU comprises 10k carefully curated audio clips paired with human-annotated natural language questions and answers spanning speech, environmental sounds, and music. It includes information extraction and reasoning questions, requiring models to demonstrate 27 distinct skills across unique and challenging tasks. Unlike existing benchmarks, MMAU emphasizes advanced perception and reasoning with domain-specific knowledge, challenging models to tackle tasks akin to those faced by experts. We assess 18 open-source and proprietary (Large) Audio-Language Models, demonstrating the significant challenges posed by MMAU. Notably, even the most advanced Gemini Pro v1.5 achieves only 52.97% accuracy, and the state-of-the-art open-source Qwen2-Audio achieves only 52.50%, highlighting considerable room for improvement. We believe MMAU will drive the audio and multimodal research community to develop more advanced audio understanding models capable of solving complex audio tasks.

Data Augmentation of Multivariate Sensor Time Series using Autoregressive Models and Application to Failure Prognostics

This work presents a novel data augmentation solution for non-stationary multivariate time series and its application to failure prognostics. The method extends previous work from the authors which is based on time-varying autoregressive processes. It can be employed to extract key information from a limited number of samples and generate new synthetic samples in a way that potentially improves the performance of PHM solutions. This is especially valuable in situations of data scarcity which are very usual in PHM, especially for failure prognostics. The proposed approach is tested based on the CMAPSS dataset, commonly employed for prognostics experiments and benchmarks. An AutoML approach from PHM literature is employed for automating the design of the prognostics solution. The empirical evaluation provides evidence that the proposed method can substantially improve the performance of PHM solutions.

SFB-net for cardiac segmentation: Bridging the semantic gap with attention

In the past few years, deep learning algorithms have been widely used for cardiac image segmentation. However, most of these architectures rely on convolutions that hardly model long-range dependencies, limiting their ability to extract contextual information. In order to tackle this issue, this article introduces the Swin Filtering Block network (SFB-net) which takes advantage of both conventional and swin transformer layers. The former are used to introduce spatial attention at the bottom of the network, while the latter are applied to focus on high level semantically rich features between the encoder and decoder. An average Dice score of 92.4 was achieved on the ACDC dataset. To the best of our knowledge, this result outperforms any other work on this dataset. The average Dice score of 87.99 obtained on the M\&M's dataset demonstrates that the proposed method generalizes well to data from different vendors and centres.

A Survey on Speech Large Language Models

Large Language Models (LLMs) exhibit strong contextual understanding and remarkable multi-task performance. Therefore, researchers have been seeking to integrate LLMs in the broad sense of Spoken Language Understanding (SLU) field. Different from the traditional method of cascading LLMs to process text generated by Automatic Speech Recognition(ASR), new efforts have focused on designing architectures centered around Audio Feature Extraction - Multimodal Information Fusion - LLM Inference(Speech LLMs). This approach enables richer audio feature extraction while simultaneously facilitating end-to-end fusion of audio and text modalities, thereby achieving deeper understanding and reasoning from audio data. This paper elucidates the development of Speech LLMs, offering an in-depth analysis of system architectures and training strategies. Through extensive research and a series of targeted experiments, the paper assesses Speech LLMs' advancements in Rich Audio Transcription and its potential for Cross-task Integration within the SLU field. Additionally, it indicates key challenges uncovered through experimentation, such as the Dormancy of LLMs under certain conditions. The paper further delves into the training strategies for Speech LLMs, proposing potential solutions based on these findings, and offering valuable insights and references for future research in this domain, as well as LLM applications in multimodal contexts.

SPEED++: A Multilingual Event Extraction Framework for Epidemic Prediction and Preparedness

Social media is often the first place where communities discuss the latest societal trends. Prior works have utilized this platform to extract epidemic-related information (e.g. infections, preventive measures) to provide early warnings for epidemic prediction. However, these works only focused on English posts, while epidemics can occur anywhere in the world, and early discussions are often in the local, non-English languages. In this work, we introduce the first multilingual Event Extraction (EE) framework SPEED++ for extracting epidemic event information for a wide range of diseases and languages. To this end, we extend a previous epidemic ontology with 20 argument roles; and curate our multilingual EE dataset SPEED++ comprising 5.1K tweets in four languages for four diseases. Annotating data in every language is infeasible; thus we develop zero-shot cross-lingual cross-disease models (i.e., training only on English COVID data) utilizing multilingual pre-training and show their efficacy in extracting epidemic-related events for 65 diverse languages across different diseases. Experiments demonstrate that our framework can provide epidemic warnings for COVID-19 in its earliest stages in Dec 2019 (3 weeks before global discussions) from Chinese Weibo posts without any training in Chinese. Furthermore, we exploit our framework's argument extraction capabilities to aggregate community epidemic discussions like symptoms and cure measures, aiding misinformation detection and public attention monitoring. Overall, we lay a strong foundation for multilingual epidemic preparedness.

BoostAdapter: Improving Vision-Language Test-Time Adaptation via Regional Bootstrapping

Adaptation of pretrained vision-language models such as CLIP to various downstream tasks have raised great interest in recent researches. Previous works have proposed a variety of test-time adaptation (TTA) methods to achieve strong generalization without any knowledge of the target domain. However, existing training-required TTA approaches like TPT necessitate entropy minimization that involves large computational overhead, while training-free methods like TDA overlook the potential for information mining from the test samples themselves. In this paper, we break down the design of existing popular training-required and training-free TTA methods and bridge the gap between them within our framework. Specifically, we maintain a light-weight key-value memory for feature retrieval from instance-agnostic historical samples and instance-aware boosting samples. The historical samples are filtered from the testing data stream and serve to extract useful information from the target distribution, while the boosting samples are drawn from regional bootstrapping and capture the knowledge of the test sample itself. We theoretically justify the rationality behind our method and empirically verify its effectiveness on both the out-of-distribution and the cross-domain datasets, showcasing its applicability in real-world situations.

DeCoRe: Decoding by Contrasting Retrieval Heads to Mitigate Hallucinations

Large Language Models (LLMs) often hallucinate, producing unfaithful or factually incorrect outputs by misrepresenting the provided context or incorrectly recalling internal knowledge. Recent studies have identified specific attention heads within the Transformer architecture, known as retrieval heads, responsible for extracting relevant contextual information. We hypothesise that masking these retrieval heads can induce hallucinations and that contrasting the outputs of the base LLM and the masked LLM can reduce hallucinations. To this end, we propose Decoding by Contrasting Retrieval Heads (DeCoRe), a novel training-free decoding strategy that amplifies information found in the context and model parameters. DeCoRe mitigates potentially hallucinated responses by dynamically contrasting the outputs of the base LLM and the masked LLM, using conditional entropy as a guide. Our extensive experiments confirm that DeCoRe significantly improves performance on tasks requiring high contextual faithfulness, such as summarisation (XSum by 18.6%), instruction following (MemoTrap by 10.9%), and open-book question answering (NQ-Open by 2.4% and NQ-Swap by 5.5%).