Establishing Task Scaling Laws via Compute-Efficient Model Ladders

We develop task scaling laws and model ladders to predict the individual task performance of pretrained language models (LMs) in the overtrained setting. Standard power laws for language modeling loss cannot accurately model task performance. Therefore, we leverage a two-step prediction approach: first use model and data size to predict a task-specific loss, and then use this task loss to predict task performance. We train a set of small-scale "ladder" models, collect data points to fit the parameterized functions of the two prediction steps, and make predictions for two target models: a 7B model trained to 4T tokens and a 13B model trained to 5T tokens. Training the ladder models only costs 1% of the compute used for the target models. On four multiple-choice tasks written in ranked classification format, we can predict the accuracy of both target models within 2 points of absolute error. We have higher prediction error on four other tasks (average absolute error 6.9) and find that these are often tasks with higher variance in task metrics. We also find that using less compute to train fewer ladder models tends to deteriorate predictions. Finally, we empirically show that our design choices and the two-step approach lead to superior performance in establishing scaling laws.

Enhancing In-Hospital Mortality Prediction Using Multi-Representational Learning with LLM-Generated Expert Summaries

In-hospital mortality (IHM) prediction for ICU patients is critical for timely interventions and efficient resource allocation. While structured physiological data provides quantitative insights, clinical notes offer unstructured, context-rich narratives. This study integrates these modalities with Large Language Model (LLM)-generated expert summaries to improve IHM prediction accuracy. Using the MIMIC-III database, we analyzed time-series physiological data and clinical notes from the first 48 hours of ICU admission. Clinical notes were concatenated chronologically for each patient and transformed into expert summaries using Med42-v2 70B. A multi-representational learning framework was developed to integrate these data sources, leveraging LLMs to enhance textual data while mitigating direct reliance on LLM predictions, which can introduce challenges in uncertainty quantification and interpretability. The proposed model achieved an AUPRC of 0.6156 (+36.41%) and an AUROC of 0.8955 (+7.64%) compared to a time-series-only baseline. Expert summaries outperformed clinical notes or time-series data alone, demonstrating the value of LLM-generated knowledge. Performance gains were consistent across demographic groups, with notable improvements in underrepresented populations, underscoring the framework's equitable application potential. By integrating LLM-generated summaries with structured and unstructured data, the framework captures complementary patient information, significantly improving predictive performance. This approach showcases the potential of LLMs to augment critical care prediction models, emphasizing the need for domain-specific validation and advanced integration strategies for broader clinical adoption.

HOBBIT: A Mixed Precision Expert Offloading System for Fast MoE Inference

The Mixture-of-Experts (MoE) architecture has demonstrated significant advantages in the era of Large Language Models (LLMs), offering enhanced capabilities with reduced inference costs. However, deploying MoE-based LLMs on memoryconstrained edge devices remains challenging due to their substantial memory requirements. While existing expertoffloading methods alleviate the memory requirements, they often incur significant expert-loading costs or compromise model accuracy. We present HOBBIT, a mixed precision expert offloading system to enable flexible and efficient MoE inference. Our key insight is that dynamically replacing less critical cache-miss experts with low precision versions can substantially reduce expert-loading latency while preserving model accuracy. HOBBIT introduces three innovative techniques that map the natural hierarchy of MoE computation: (1) a token-level dynamic expert loading mechanism, (2) a layer-level adaptive expert prefetching technique, and (3) a sequence-level multidimensional expert caching policy. These innovations fully leverage the benefits of mixedprecision expert inference. By implementing HOBBIT on top of the renowned LLM inference framework Llama.cpp, we evaluate its performance across different edge devices with representative MoE models. The results demonstrate that HOBBIT achieves up to a 9.93x speedup in decoding compared to state-of-the-art MoE offloading systems.

AI as Humanity's Salieri: Quantifying Linguistic Creativity of Language Models via Systematic Attribution of Machine Text against Web Text

Creativity has long been considered one of the most difficult aspect of human intelligence for AI to mimic. However, the rise of Large Language Models (LLMs), like ChatGPT, has raised questions about whether AI can match or even surpass human creativity. We present CREATIVITY INDEX as the first step to quantify the linguistic creativity of a text by reconstructing it from existing text snippets on the web. CREATIVITY INDEX is motivated by the hypothesis that the seemingly remarkable creativity of LLMs may be attributable in large part to the creativity of human-written texts on the web. To compute CREATIVITY INDEX efficiently, we introduce DJ SEARCH, a novel dynamic programming algorithm that can search verbatim and near-verbatim matches of text snippets from a given document against the web. Experiments reveal that the CREATIVITY INDEX of professional human authors is on average 66.2% higher than that of LLMs, and that alignment reduces the CREATIVITY INDEX of LLMs by an average of 30.1%. In addition, we find that distinguished authors like Hemingway exhibit measurably higher CREATIVITY INDEX compared to other human writers. Finally, we demonstrate that CREATIVITY INDEX can be used as a surprisingly effective criterion for zero-shot machine text detection, surpassing the strongest existing zero-shot system, DetectGPT, by a significant margin of 30.2%, and even outperforming the strongest supervised system, GhostBuster, in five out of six domains.

Unpacking DPO and PPO: Disentangling Best Practices for Learning from Preference Feedback

Learning from preference feedback has emerged as an essential step for improving the generation quality and performance of modern language models (LMs). Despite its widespread use, the way preference-based learning is applied varies wildly, with differing data, learning algorithms, and evaluations used, making disentangling the impact of each aspect difficult. In this work, we identify four core aspects of preference-based learning: preference data, learning algorithm, reward model, and policy training prompts, systematically investigate the impact of these components on downstream model performance, and suggest a recipe for strong learning for preference feedback. Our findings indicate that all aspects are important for performance, with better preference data leading to the largest improvements, followed by the choice of learning algorithm, the use of improved reward models, and finally the use of additional unlabeled prompts for policy training. Notably, PPO outperforms DPO by up to 2.5% in math and 1.2% in general domains. High-quality preference data leads to improvements of up to 8% in instruction following and truthfulness. Despite significant gains of up to 5% in mathematical evaluation when scaling up reward models, we surprisingly observe marginal improvements in other categories. We publicly release the code used for training (https://github.com/hamishivi/EasyLM) and evaluating (https://github.com/allenai/open-instruct) our models, along with the models and datasets themselves (https://huggingface.co/collections/allenai/tulu-v25-suite-66676520fd578080e126f618).

DiffPop: Plausibility-Guided Object Placement Diffusion for Image Composition

In this paper, we address the problem of plausible object placement for the challenging task of realistic image composition. We propose DiffPop, the first framework that utilizes plausibility-guided denoising diffusion probabilistic model to learn the scale and spatial relations among multiple objects and the corresponding scene image. First, we train an unguided diffusion model to directly learn the object placement parameters in a self-supervised manner. Then, we develop a human-in-the-loop pipeline which exploits human labeling on the diffusion-generated composite images to provide the weak supervision for training a structural plausibility classifier. The classifier is further used to guide the diffusion sampling process towards generating the plausible object placement. Experimental results verify the superiority of our method for producing plausible and diverse composite images on the new Cityscapes-OP dataset and the public OPA dataset, as well as demonstrate its potential in applications such as data augmentation and multi-object placement tasks. Our dataset and code will be released.

Camera-Based Remote Physiology Sensing for Hundreds of Subjects Across Skin Tones

Remote photoplethysmography (rPPG) emerges as a promising method for non-invasive, convenient measurement of vital signs, utilizing the widespread presence of cameras. Despite advancements, existing datasets fall short in terms of size and diversity, limiting comprehensive evaluation under diverse conditions. This paper presents an in-depth analysis of the VitalVideo dataset, the largest real-world rPPG dataset to date, encompassing 893 subjects and 6 Fitzpatrick skin tones. Our experimentation with six unsupervised methods and three supervised models demonstrates that datasets comprising a few hundred subjects(i.e., 300 for UBFC-rPPG, 500 for PURE, and 700 for MMPD-Simple) are sufficient for effective rPPG model training. Our findings highlight the importance of diversity and consistency in skin tones for precise performance evaluation across different datasets.

Terahertz channel modeling based on surface sensing characteristics

The dielectric properties of environmental surfaces, including walls, floors and the ground, etc., play a crucial role in shaping the accuracy of terahertz (THz) channel modeling, thereby directly impacting the effectiveness of communication systems. Traditionally, acquiring these properties has relied on methods such as terahertz time-domain spectroscopy (THz-TDS) or vector network analyzers (VNA), demanding rigorous sample preparation and entailing a significant expenditure of time. However, such measurements are not always feasible, particularly in novel and uncharacterized scenarios. In this work, we propose a new approach for channel modeling that leverages the inherent sensing capabilities of THz channels. By comparing the results obtained through channel sensing with that derived from THz-TDS measurements, we demonstrate the method's ability to yield dependable surface property information. The application of this approach in both a miniaturized cityscape scenario and an indoor environment has shown consistency with experimental measurements, thereby verifying its effectiveness in real-world settings.

Ground-to-UAV 140 GHz channel measurement and modeling

Unmanned Aerial Vehicle (UAV) assisted terahertz (THz) wireless communications have been expected to play a vital role in the next generation of wireless networks. UAVs can serve as either repeaters or data collectors within the communication link, thereby potentially augmenting the efficacy of communication systems. Despite their promise, the channel analysis and modeling specific to THz wireless channels leveraging UAVs remain under explored. This work delves into a ground-to-UAV channel at 140 GHz, with a specific focus on the influence of UAV hovering behavior on channel performance. Employing experimental measurements through an unmodulated channel setup and a geometry-based stochastic model (GBSM) that integrates three-dimensional positional coordinates and beamwidth, this work evaluates the impact of UAV dynamic movements and antenna orientation on channel performance. Our findings highlight the minimal impact of UAV orientation adjustments on channel performance and underscore the diminishing necessity for precise alignment between UAVs and ground stations as beamwidth increases.

An Interpretable Power System Transient Stability Assessment Method with Expert Guiding Neural-Regression-Tree

Deep learning based transient stability assessment (TSA) has achieved great success, yet the lack of interpretability hinders its industrial application. Although a great number of studies have tried to explore the interpretability of network solutions, many problems still remain unsolved: (1) the difference between the widely accepted power system knowledge and the generated interpretive rules is large, (2) the probability characteristics of the neural network have not been fully considered during generating the interpretive rules, (3) the cost of the trade-off between accuracy and interpretability is too heavy to take. To address these issues, an interpretable power system Transient Stability Assessment method with Expert guiding Neural-Regression-Tree (TSA-ENRT) is proposed. TSA-ENRT utilizes an expert guiding nonlinear regression tree to approximate the neural network prediction and the neural network can be explained by the interpretive rules generated by the tree model. The nonlinearity of the expert guiding nonlinear regression tree is endowed with the extracted knowledge from a simple two-machine three-bus power system, which forms an expert knowledge base and thus the generated interpretive rules are more consistent with human cognition. Besides, the expert guiding tree model can build a bridge between the interpretive rules and the probability prediction of neural network in a regression way. By regularizing the neural network with the average decision length of ENRT, the association of the neural network and tree model is constructed in the model training level which provides a better trade-off between accuracy and interpretability. Extensive experiments indicate the interpretive rules generated by the proposed TSA-ENRT are highly consistent with the neural network prediction and more agreed with human expert cognition.