Retrieval-augmented generation (RAG) methods are viable solutions for addressing the static memory limits of pre-trained language models. Nevertheless, encountering conflicting sources of information within the retrieval context is an inevitable practical challenge. In such situations, the language models are recommended to transparently inform users about the conflicts rather than autonomously deciding what to present based on their inherent biases. To analyze how current large language models (LLMs) align with our recommendation, we introduce WhoQA, a public benchmark dataset to examine model's behavior in knowledge conflict situations. We induce conflicts by asking about a common property among entities having the same name, resulting in questions with up to 8 distinctive answers. WhoQA evaluation set includes 5K questions across 13 Wikidata property types and 150K Wikipedia entities. Our experiments show that despite the simplicity of WhoQA questions, knowledge conflicts significantly degrades LLMs' performance in RAG settings.

The impact of voice disorders is becoming more widely acknowledged as a public health issue. Several machine learning-based classifiers with the potential to identify disorders have been used in recent studies to differentiate between normal and pathological voices and sounds. In this paper, we focus on analyzing the vulnerabilities of these systems by exploring the possibility of attacks that can reverse classification and compromise their reliability. Given the critical nature of personal health information, understanding which types of attacks are effective is a necessary first step toward improving the security of such systems. Starting from the original audios, we implement various attack methods, including adversarial, evasion, and pitching techniques, and evaluate how state-of-the-art disorder detection models respond to them. Our findings identify the most effective attack strategies, underscoring the need to address these vulnerabilities in machine-learning systems used in the healthcare domain.

Classic tree-based ensembles generalize better than any single decision tree. In contrast, recent empirical studies find that modern ensembles of (overparameterized) neural networks may not provide any inherent generalization advantage over single but larger neural networks. This paper clarifies how modern overparameterized ensembles differ from their classic underparameterized counterparts, using ensembles of random feature (RF) regressors as a basis for developing theory. In contrast to the underparameterized regime, where ensembling typically induces regularization and increases generalization, we prove that infinite ensembles of overparameterized RF regressors become pointwise equivalent to (single) infinite-width RF regressors. This equivalence, which is exact for ridgeless models and approximate for small ridge penalties, implies that overparameterized ensembles and single large models exhibit nearly identical generalization. As a consequence, we can characterize the predictive variance amongst ensemble members, and demonstrate that it quantifies the expected effects of increasing capacity rather than capturing any conventional notion of uncertainty. Our results challenge common assumptions about the advantages of ensembles in overparameterized settings, prompting a reconsideration of how well intuitions from underparameterized ensembles transfer to deep ensembles and the overparameterized regime. Ensembling is one of the most well-established techniques in machine learning (e.g. Historically, most ensembles aggregated component models that are simple by today's standards.

Bayesian inference offers a robust framework for updating prior beliefs based on new data using Bayes' theorem, but exact inference is often computationally infeasible, necessitating approximate methods. Though widely used, these methods struggle to estimate marginal likelihoods accurately, particularly due to the rigid functional structures of deterministic models like Gaussian processes and the limitations of small sample sizes in stochastic models like the ensemble Kalman method. In this work, we introduce BI-EqNO, an equivariant neural operator framework for generalized approximate Bayesian inference, designed to enhance both deterministic and stochastic approaches. BI-EqNO transforms priors into posteriors conditioned on observation data through data-driven training. The framework is flexible, supporting diverse prior and posterior representations with arbitrary discretizations and varying numbers of observations. Crucially, BI-EqNO's architecture ensures (1) permutation equivariance between prior and posterior representations, and (2) permutation invariance with respect to observational data. We demonstrate BI-EqNO's utility through two examples: (1) as a generalized Gaussian process (gGP) for regression, and (2) as an ensemble neural filter (EnNF) for sequential data assimilation. Results show that gGP outperforms traditional Gaussian processes by offering a more flexible representation of covariance functions. Additionally, EnNF not only outperforms the ensemble Kalman filter in small-ensemble settings but also has the potential to function as a "super" ensemble filter, capable of representing and integrating multiple ensemble filters for enhanced assimilation performance. This study highlights BI-EqNO's versatility and effectiveness, improving Bayesian inference through data-driven training while reducing computational costs across various applications.

Reward models are critical in techniques like Reinforcement Learning from Human Feedback (RLHF) and Inference Scaling Laws, where they guide language model alignment and select optimal responses. Despite their importance, existing reward model benchmarks often evaluate models by asking them to distinguish between responses generated by models of varying power. However, this approach fails to assess reward models on subtle but critical content changes and variations in style, resulting in a low correlation with policy model performance. Reward models play a pivotal role in both techniques. In RLHF, reward models serve as proxies for human values, providing feedback on generated text, which helps align language models (policy models) during training (Ouyang et al., 2022; Dong et al., 2024). In Inference Scaling Law, reward models are used to select the best response from a set of candidates based on predicted rewards (Wu et al., 2024; Snell et al., 2024). Despite their significance, benchmarks for reward models remain under-explored compared to the rapid advancements in aligned language model evaluation, namely the policy model (Hendrycks et al., 2020; bench authors, 2023; Chiang et al., 2024; Hendrycks et al., 2021). To conduct a faithful and systematical evaluation, an ideal benchmark for reward models should adhere to three key principles: 1) Assessing Reward Models' Sensitivity to Subtle Changes: A faithful reward model should sensitively distinguish subtle changes and assign a higher reward to the correct response. For example, in Table 1, Response 1 and Response 2 differ by only one word but express completely different meanings, requiring the reward model to focus on content quality. For example, in Table 1, Response 3 is factually incorrect but longer than Response 1, which could mislead the reward model into assigning a higher reward to Response 3. 3) Correlating with Policy Models: A good reward model benchmark should highly correlate with the performance of the aligned language model (the policy model). This would make it a reliable proxy for selecting the best reward model for alignment. Recent efforts (Lambert et al., 2024; Zhu et al., 2023; Jiang et al., 2023) have made progress by constructing benchmarks from existing preference datasets.

Fine-tuning language models has become increasingly popular following the proliferation of open models and improvements in cost-effective parameter efficient fine-tuning. However, fine-tuning can influence model properties such as safety. We assess how fine-tuning can impact different open models' propensity to output toxic content. We assess the impacts of fine-tuning Gemma, Llama, and Phi models on toxicity through three experiments. We compare how toxicity is reduced by model developers during instruction-tuning. We show that small amounts of parameter-efficient fine-tuning on developer-tuned models via low-rank adaptation on a non-adversarial dataset can significantly alter these results across models. Finally, we highlight the impact of this in the wild, demonstrating how toxicity rates of models fine-tuned by community contributors can deviate in hard-to-predict ways.

The integration of Autonomous Weapon Systems (AWS) into military operations presents both significant opportunities and challenges. This paper explores the multifaceted nature of trust in AWS, emphasising the necessity of establishing reliable and transparent systems to mitigate risks associated with bias, operational failures, and accountability. Despite advancements in Artificial Intelligence (AI), the trustworthiness of these systems, especially in high-stakes military applications, remains a critical issue. Through a systematic review of existing literature, this research identifies gaps in the understanding of trust dynamics during the development and deployment phases of AWS. It advocates for a collaborative approach that includes technologists, ethicists, and military strategists to address these ongoing challenges. The findings underscore the importance of Human-Machine teaming and enhancing system intelligibility to ensure accountability and adherence to International Humanitarian Law. Ultimately, this paper aims to contribute to the ongoing discourse on the ethical implications of AWS and the imperative for trustworthy AI in defense contexts.

Financial shocks, such as the one observed during the global financial crisis, exhibit important domestic and international consequences on macroeconomic aggregates (see, e.g., Dovern and van Roye, 2014; Ciccarelli et al., 2016; Prieto et al., 2016; Gerba et al., 2024). Policymakers in central banks and governmental institutions, who aim to smooth business cycles and thus alleviate the negative effects of adverse financial disruptions, need to understand how such shocks impact the economy and propagate internationally to implement policies in a forward-looking manner. The recent literature provides plenty of evidence on the domestic and international effects of US financial shocks (see Balke, 2000; Gilchrist and Zakrajšek, 2012; Cesa-Bianchi and Sokol, 2022). These papers find that financial shocks exert powerful effects on domestic output but also that US-based shocks spill over to foreign economies and trigger declines in international economic activity. Such effects might be subject to time variation (Abbate et al., 2016).

Modern network defense can benefit from the use of autonomous systems, offloading tedious and time-consuming work to agents with standard and learning-enabled components. These agents, operating on critical network infrastructure, need to be robust and trustworthy to ensure defense against adaptive cyber-attackers and, simultaneously, provide explanations for their actions and network activity. However, learning-enabled components typically use models, such as deep neural networks, that are not transparent in their high-level decision-making leading to assurance challenges. Additionally, cyber-defense agents must execute complex long-term defense tasks in a reactive manner that involve coordination of multiple interdependent subtasks. Behavior trees are known to be successful in modelling interpretable, reactive, and modular agent policies with learning-enabled components. In this paper, we develop an approach to design autonomous cyber defense agents using behavior trees with learning-enabled components, which we refer to as Evolving Behavior Trees (EBTs). We learn the structure of an EBT with a novel abstract cyber environment and optimize learning-enabled components for deployment. The learning-enabled components are optimized for adapting to various cyber-attacks and deploying security mechanisms. The learned EBT structure is evaluated in a simulated cyber environment, where it effectively mitigates threats and enhances network visibility. For deployment, we develop a software architecture for evaluating EBT-based agents in computer network defense scenarios. Our results demonstrate that the EBT-based agent is robust to adaptive cyber-attacks and provides high-level explanations for interpreting its decisions and actions.

We present efforts in the fields of machine learning and time series forecasting to accurately predict counts of future opioid overdose incidents recorded by Emergency Medical Services (EMS) in the state of Kentucky. Forecasts are useful to state government agencies to properly prepare and distribute resources related to opioid overdoses effectively. Our approach uses county and district level aggregations of EMS opioid overdose encounters and forecasts future counts for each month. A variety of additional covariates were tested to determine their impact on the model's performance. Models with different levels of complexity were evaluated to optimize training time and accuracy. Our results show that when special precautions are taken to address data sparsity, useful predictions can be generated with limited error by utilizing yearly trends and covariance with additional data sources.