Reward models (RMs) are at the crux of successfully using RLHF to align pretrained models to human preferences, yet there has been relatively little study that focuses on evaluation of those models. Evaluating reward models presents an opportunity to understand the opaque technologies used for alignment of language models and which values are embedded in them. Resources for reward model training and understanding are sparse in the nascent open-source community around them. To enhance scientific understanding of reward models, we present RewardBench, a benchmark dataset and code-base for evaluation. The RewardBench dataset is a collection of prompt-chosen-rejected trios spanning chat, reasoning, and safety, to benchmark how reward models perform on challenging, structured and out-of-distribution queries. We create specific comparison datasets for RMs that have subtle, but verifiable reasons (e.g. bugs, incorrect facts) why one answer should be preferred to another. On the RewardBench leaderboard, we evaluate reward models trained with a variety of methods, such as the direct MLE training of classifiers and the implicit reward modeling of Direct Preference Optimization (DPO). We present many findings on propensity for refusals, reasoning limitations, and instruction following shortcomings of various reward models towards a better understanding of the RLHF process.

Accurate time series forecasts are crucial for various applications, such as traffic management, electricity consumption, and healthcare. However, limitations in models and data quality can significantly impact forecasts accuracy. One common issue with data quality is the absence of data points, referred to as missing data. It is often caused by sensor malfunctions, equipment failures, or human errors. This paper proposes Hinge-FM2I, a novel method for handling missing data values in univariate time series data. Hinge-FM2I builds upon the strengths of the Forecasting Method by Image Inpainting (FM2I). FM2I has proven effective, but selecting the most accurate forecasts remain a challenge. To overcome this issue, we proposed a selection algorithm. Inspired by door hinges, Hinge-FM2I drops a data point either before or after the gap (left/right-hinge), then use FM2I for imputation, and then select the imputed gap based on the lowest error of the dropped data point. Hinge-FM2I was evaluated on a comprehensive sample composed of 1356 time series, extracted from the M3 competition benchmark dataset, with missing value rates ranging from 3.57\% to 28.57\%. Experimental results demonstrate that Hinge-FM2I significantly outperforms established methods such as, linear/spline interpolation, K-Nearest Neighbors (K-NN), and ARIMA. Notably, Hinge-FM2I achieves an average Symmetric Mean Absolute Percentage Error (sMAPE) score of 5.6\% for small gaps, and up to 10\% for larger ones. These findings highlight the effectiveness of Hinge-FM2I as a promising new method for addressing missing values in univariate time series data.

The integration of generative artificial intelligence (GenAI) tools into education has been a game-changer for teaching and assessment practices, bringing new opportunities, but also novel challenges which need to be dealt with. This paper presents the Comprehensive AI Assessment Framework (CAIAF), an evolved version of the AI Assessment Scale (AIAS) by Perkins, Furze, Roe, and MacVaugh, targeted toward the ethical integration of AI into educational assessments. This is where the CAIAF differs, as it incorporates stringent ethical guidelines, with clear distinctions based on educational levels, and advanced AI capabilities of real-time interactions and personalized assistance. The framework developed herein has a very intuitive use, mainly through the use of a color gradient that enhances the user-friendliness of the framework. Methodologically, the framework has been developed through the huge support of a thorough literature review and practical insight into the topic, becoming a dynamic tool to be used in different educational settings. The framework will ensure better learning outcomes, uphold academic integrity, and promote responsible use of AI, hence the need for this framework in modern educational practice.

Retrieval Augmented Generation (RAG) enhances the abilities of Large Language Models (LLMs) by enabling the retrieval of documents into the LLM context to provide more accurate and relevant responses. Existing RAG solutions do not focus on queries that may require fetching multiple documents with substantially different contents. Such queries occur frequently, but are challenging because the embeddings of these documents may be distant in the embedding space, making it hard to retrieve them all. This paper introduces Multi-Head RAG (MRAG), a novel scheme designed to address this gap with a simple yet powerful idea: leveraging activations of Transformer's multi-head attention layer, instead of the decoder layer, as keys for fetching multi-aspect documents. The driving motivation is that different attention heads can learn to capture different data aspects. Harnessing the corresponding activations results in embeddings that represent various facets of data items and queries, improving the retrieval accuracy for complex queries. We provide an evaluation methodology and metrics, synthetic datasets, and real-world use cases to demonstrate MRAG's effectiveness, showing improvements of up to 20% in relevance over standard RAG baselines. MRAG can be seamlessly integrated with existing RAG frameworks and benchmarking tools like RAGAS as well as different classes of data stores.

There is an increase in the proliferation of online hate commensurate with the rise in the usage of social media. In response, there is also a significant advancement in the creation of automated tools aimed at identifying harmful text content using approaches grounded in Natural Language Processing and Deep Learning. Although it is known that training Deep Learning models require a substantial amount of annotated data, recent line of work suggests that models trained on specific subsets of the data still retain performance comparable to the model that was trained on the full dataset. In this work, we show how we can leverage influence scores to estimate the importance of a data point while training a model and designing a pruning strategy applied to the case of sexism detection. We evaluate the model performance trained on data pruned with different pruning strategies on three out-of-domain datasets and find, that in accordance with other work a large fraction of instances can be removed without significant performance drop. However, we also discover that the strategies for pruning data, previously successful in Natural Language Inference tasks, do not readily apply to the detection of harmful content and instead amplify the already prevalent class imbalance even more, leading in the worst-case to a complete absence of the hateful class.

Generative AI (GenAI) will have significant impact on content creation platforms. In this paper, we study the dynamic competition between a GenAI and a human contributor. Unlike the human, the GenAI's content only improves when more contents are created by human over the time; however, GenAI has the advantage of generating content at a lower cost. We study the algorithmic problem in this dynamic competition model about how the human contributor can maximize her utility when competing against the GenAI for content generation over a set of topics. In time-sensitive content domains (e.g., news or pop music creation) where contents' value diminishes over time, we show that there is no polynomial time algorithm for finding the human's optimal (dynamic) strategy, unless the randomized exponential time hypothesis is false. Fortunately, we are able to design a polynomial time algorithm that naturally cycles between myopically optimizing over a short time window and pausing and provably guarantees an approximation ratio of $\frac{1}{2}$. We then turn to time-insensitive content domains where contents do not lose their value (e.g., contents on history facts). Interestingly, we show that this setting permits a polynomial time algorithm that maximizes the human's utility in the long run.

This study investigates pedestrian crash severity through Automated Machine Learning (AutoML), offering a streamlined and accessible method for analyzing critical factors. Utilizing a detailed dataset from Utah spanning 2010-2021, the research employs AutoML to assess the effects of various explanatory variables on crash outcomes. The study incorporates SHAP (SHapley Additive exPlanations) to interpret the contributions of individual features in the predictive model, enhancing the understanding of influential factors such as lighting conditions, road type, and weather on pedestrian crash severity. Emphasizing the efficiency and democratization of data-driven methodologies, the paper discusses the benefits of using AutoML in traffic safety analysis. This integration of AutoML with SHAP analysis not only bolsters predictive accuracy but also improves interpretability, offering critical insights into effective pedestrian safety measures. The findings highlight the potential of this approach in advancing the analysis of pedestrian crash severity.

Deep neural networks (DNNs) are one of the most widely used machine learning algorithm. DNNs requires the training data to be available beforehand with true labels. This is not feasible for many real-world problems where data arrives in the streaming form and acquisition of true labels are scarce and expensive. In the literature, not much focus has been given to the privacy prospect of the streaming data, where data may change its distribution frequently. These concept drifts must be detected privately in order to avoid any disclosure risk from DNNs. Existing privacy models use concept drift detection schemes such ADWIN, KSWIN to detect the drifts. In this paper, we focus on the notion of integrally private DNNs to detect concept drifts. Integrally private DNNs are the models which recur frequently from different datasets. Based on this, we introduce an ensemble methodology which we call 'Integrally Private Drift Detection' (IPDD) method to detect concept drift from private models. Our IPDD method does not require labels to detect drift but assumes true labels are available once the drift has been detected. We have experimented with binary and multi-class synthetic and real-world data. Our experimental results show that our methodology can privately detect concept drift, has comparable utility (even better in some cases) with ADWIN and outperforms utility from different levels of differentially private models. The source code for the paper is available \hyperlink{https://github.com/Ayush-Umu/Concept-drift-detection-Using-Integrally-private-models}{here}.

Field-of-view (FOV) recovery of truncated chest CT scans is crucial for accurate body composition analysis, which involves quantifying skeletal muscle and subcutaneous adipose tissue (SAT) on CT slices. This, in turn, enables disease prognostication. Here, we present a method for recovering truncated CT slices using generative image outpainting. We train a diffusion model and apply it to truncated CT slices generated by simulating a small FOV. Our model reliably recovers the truncated anatomy and outperforms the previous state-of-the-art despite being trained on 87% less data.

Selective classification enables models to make predictions only when they are sufficiently confident, aiming to enhance safety and reliability, which is important in high-stakes scenarios. Previous methods mainly use deep neural networks and focus on modifying the architecture of classification layers to enable the model to estimate the confidence of its prediction. This work provides a generalization bound for selective classification, disclosing that optimizing feature layers helps improve the performance of selective classification. Inspired by this theory, we propose to explicitly improve the selective classification model at the feature level for the first time, leading to a novel Confidence-aware Contrastive Learning method for Selective Classification, CCL-SC, which similarizes the features of homogeneous instances and differentiates the features of heterogeneous instances, with the strength controlled by the model's confidence. The experimental results on typical datasets, i.e., CIFAR-10, CIFAR-100, CelebA, and ImageNet, show that CCL-SC achieves significantly lower selective risk than state-of-the-art methods, across almost all coverage degrees. Moreover, it can be combined with existing methods to bring further improvement.