Study evaluates new models using human feedback from another model.
problem Evaluate a new model using human feedback collected for another model.
method Formalize problem, propose model-based and model-free estimators, analyze unbiasedness, and empirically evaluate.
result Proposed estimators can predict absolute values, rank, and optimize evaluated policies.
We conduct a large-scale, systematic study to evaluate the existing evaluation methods for natural language generation in the context of generating online product reviews. We compare human-based evaluators with a variety of automated evaluation procedures, including discriminative evaluators that measure how well machi…
This paper proposes a method to use LLMs as auxiliary evaluators in place of human judges.
problem The need for cost-effective and scalable evaluation of AI systems.
method Formulates a two-stage sampling design with LLM evaluations and human ratings, using a doubly robust estimator.
result Proposes a method to determine optimal sample sizes for human and LLM ratings.
How can we measure whether a natural language generation system produces both high quality and diverse outputs? Human evaluation captures quality but not diversity, as it does not catch models that simply plagiarize from the training set. On the other hand, statistical evaluation (i.e., perplexity) captures diversity b…
With success on controlled tasks, generative models are being increasingly applied to humanitarian applications [1,2]. In this paper, we focus on the evaluation of a conditional generative model that illustrates the consequences of climate change-induced flooding to encourage public interest and awareness on the issue.…
Generative models have made immense progress in recent years, particularly in their ability to generate high quality images. However, that quality has been difficult to evaluate rigorously, with evaluation dominated by heuristic approaches that do not correlate well with human judgment, such as the Inception Score and …
Efficiently evaluate generative models at the prompt level using tensor factorization.
problem Fine-grained evaluations of generative models are costly and often misaligned with human judgment.
method Tensor factorization model that merges cheap autorater data with a small set of human gold-standard labels.
result The method provides accurate and tight confidence intervals for model performance.
Topic models are typically evaluated with respect to the global topic distributions that they generate, using metrics such as coherence, but without regard to local (token-level) topic assignments. Token-level assignments are important for downstream tasks such as classification. Even recent models, which aim to improv…
Study compares DSPy teleprompter algorithms for aligning LLM evaluations with human annotations.
problem Aligning LLM evaluation metrics with human annotations.
method Comparative analysis of five teleprompter algorithms within the DSPy framework.
result Certain teleprompters outperform others in detecting hallucinations.
Meena is a chatbot trained on social media data, achieving human-like conversation quality.
problem Creating a chatbot that can have human-like conversations in an open-domain setting.
method End-to-end training of a 2.6B parameter neural network on social media data, using perplexity and a human evaluation metric (SSA) to assess quality.
result Meena achieves a high human-like conversation quality (79% SSA) when compared to existing chatbots.
LLM evaluation suffers from systematic biases and lacks reliable positive judgments.
problem LLM evaluation suffers from systematic biases and lacks reliable positive judgments.
method Formulate LLM evaluation as a positive-unlabelled learning problem and propose a geometric auditing framework based on Partial Optimal Transport.
result Improved alignment with human preferences, increased robustness to presentation biases, and interpretable confidence estimates.
New framework evaluates model explanations based on decision task improvement.
problem Evaluation of model explanations often misses practical value.
method Decision-theoretic framework quantifying three key values.
result Provides benchmarks and interprets human-AI decision support.
Study shows current metrics for audio adversarial examples are unreliable for human perception.
problem The reliability of metrics for evaluating audio adversarial examples.
method Analytical framework and human evaluation experiment.
result Current metrics for audio adversarial examples are not reliable for human perception.
This paper improves sample efficiency for off-policy evaluation with preference data.
problem Improving sample efficiency for off-policy evaluation with preference data.
method Using a deep neural network to learn the value function and leveraging manifold structure.
result Established a provably efficient guarantee for off-policy evaluation with RLHF.
AURA: Adaptive Uncertainty-aware Refinement for LLM-as-a-Judge Auditing
problem Auditing LLM-as-a-Judge decisions
method Adaptive uncertainty-aware refinement
result Human-consistent signal learning and evidence propagation
Experiment shows cognitive biases impact human-AI collaboration, highlighting the need for diverse evaluator samples.
problem Cognitive biases affect human-AI collaboration, leading to suboptimal outcomes.
method Randomized experiment with 2,784 participants, manipulating AI suggestion quality, task burden, and financial incentives.
result Individual attitudes toward AI are the strongest predictor of performance, influencing accuracy and overcorrection.
A test measures artificial agents' human-like behavior in video games.
problem Measuring the believability of artificial agents' human-like behavior.
method Developed a non-parametric two-sample hypothesis test.
result The p-value correlates with human judgment of human-like behavior. Our research extends the Bilingual Evaluation Understudy (BLEU) evaluation technique for statistical machine translation to make it more adjustable and robust. We intend to adapt it to resemble human evaluation more. We perform experiments to evaluate the performance of our technique against the primary existing evalua…
Unified framework to bridge human and LLM judgments.
problem Systematic discrepancies between human and LLM evaluations.
method Latent human preference score and linear transformations of covariates.
result Higher agreement with human ratings and exposure of systematic gaps.
Corrects bias in LLM-as-a-judge evaluations using adaptive calibration.
problem Bias in LLM evaluations due to imperfect sensitivity and specificity.
method Plug-in framework with confidence intervals accounting for test and calibration dataset uncertainties.
result LML-based evaluation yields more reliable estimates than human-only evaluation.
This paper evaluates deferring systems using causal inference.
problem Evaluating the impact of deferring systems on model accuracy is challenging.
method The paper uses a causal inference framework to evaluate deferring systems, distinguishing between scenarios with and without access to human predictions.
result The approach allows identifying causal effects of deferring strategies on predictive accuracy.
AI benchmarks evaluate football team performance using generative models.
problem Evaluating human performance in complex interactive tasks is error-prone and unreliable.
method Trained Conditional VRNN Model on player and ball tracking data to imitate and predict team interactions.
result Trained model as a useful benchmark for evaluating team performance in football.
Building an open-domain conversational agent is a challenging problem. Current evaluation methods, mostly post-hoc judgments of static conversation, do not capture conversation quality in a realistic interactive context. In this paper, we investigate interactive human evaluation and provide evidence for its necessity; …
Bayesian framework evaluates predictors of subjective visual tasks.
problem Evaluating uncertainty in machine learning predictors for tasks with subjective annotations.
method Bayesian framework to estimate epistemic uncertainty from human labels.
result Framework successfully applied to four image classification tasks.
Study flaws in generative model evaluation metrics, especially for diffusion models.
problem Flaws in existing metrics for evaluating generative models, particularly for diffusion models.
method Systematic study of generative models, human perception experiments, and analysis of feature extractors.
result State-of-the-art perceptual realism of diffusion models is not reflected in commonly reported metrics.
Study evaluates five LLMs for financial report analysis, revealing performance differences and variability.
problem Lack of understanding in reliability, consistency, and transparency of LLMs in financial analysis.
method Human evaluation, automated similarity metrics, and behavioral diagnostics applied to five transformer-based LLMs over U.S. 10-K filings.
result No single LLM consistently dominates across all evaluation perspectives, highlighting variability and need for interpretability.
Transparency, user trust, and human comprehension are popular ethical motivations for interpretable machine learning. In support of these goals, researchers evaluate model explanation performance using humans and real world applications. This alone presents a challenge in many areas of artificial intelligence. In this …
Owing to the advancement of deep learning, artificial systems are now rival to humans in several pattern recognition tasks, such as visual recognition of object categories. However, this is only the case with the tasks for which correct answers exist independent of human perception. There is another type of tasks for w…
This paper demonstrates the use of genetic algorithms for evolving a grandmaster-level evaluation function for a chess program. This is achieved by combining supervised and unsupervised learning. In the supervised learning phase the organisms are evolved to mimic the behavior of human grandmasters, and in the unsupervi…
UCFE benchmarks LLMs in financial tasks with human feedback.
problem Evaluating LLMs' financial task performance and user satisfaction.
method Hybrid approach combining human expert evaluations and dynamic interactions.
result Significant alignment between benchmark scores and human preferences (Pearson correlation coefficient of 0.78).
A method for trust evaluation of devices in human-device coexistence systems.
problem Efficient trust evaluation of devices in systems with diverse physical and social attributes.
method Canonical correlation analysis-enhanced hypergraph self-supervised learning (HSLCCA).
result The proposed HSLCCA method significantly outperforms baseline algorithms in identifying trusted devices.
We conduct large-scale studies on `human attention' in Visual Question Answering (VQA) to understand where humans choose to look to answer questions about images. We design and test multiple game-inspired novel attention-annotation interfaces that require the subject to sharpen regions of a blurred image to answer a qu…
RandomNet uses random search to design neural architectures without much human intervention.
problem Designing neural architectures without excessive human intervention.
method Random search strategy for multimodal neural architecture design.
result RandomNet performs close to state-of-the-art on AV-MNIST with minimal human supervision.
Robust OPE framework uses human inputs to improve policy evaluation in changing environments.
problem Inaccurate policy evaluations due to shifts in environment properties.
method Adapts OPE methods to shifts on user-inputted covariates, providing more realistic utility estimates.
result Robust OPE framework yields less pessimistic policy evaluations and captures realistic dataset shifts.
Many challenges in natural language processing require generating text, including language translation, dialogue generation, and speech recognition. For all of these problems, text generation becomes more difficult as the text becomes longer. Current language models often struggle to keep track of coherence for long pi…
Robot learns from human demonstrations to work autonomously.
problem Leveraging human and robotic strengths in human-robot systems.
method Bayesian inference for detecting and classifying human heterogeneity.
result Bayesian approach outperforms conventional methods by up to 12.8. This paper argues for more realistic human models in RL.
problem Current RL models oversimplify human feedback, ignoring personal, contextual, and dynamic aspects.
method Calls for interdisciplinary research on human feedback in RL.
result Realistic human models are needed for robust human-in-the-loop RL systems.
Study on predictable forward processes in trading without frequent evaluations.
problem Trading performance evaluation times not matching trading times.
method Solving a linear functional equation to construct predictable forward processes.
result Predictable forward processes are inherently myopic and optimal strategies do not use future information.
Study finds gender bias in human evaluators and shows how machine learning can mitigate it.
problem Gender bias in human decision-making on micro-lending platforms.
method Structural econometric model and machine learning algorithms trained on real-world data.
result Machine learning algorithms can mitigate both preference-based and belief-based biases.
When might human input help (or not) when assessing risk in fairness domains? Dressel and Farid (2018) asked Mechanical Turk workers to evaluate a subset of defendants in the ProPublica COMPAS data for risk of recidivism, and concluded that COMPAS predictions were no more accurate or fair than predictions made by human…
Grading function evaluates Bach-style chorales, outperforming human experts.
problem Difficulty in automatically evaluating musical style correctness.
method Introduces a grading function for evaluating four-part chorales in the style of J.S. Bach.
result Transformer model output is outperformed by the grading function at discriminating Bach chorales.
Evaluates change point detection algorithms on real-world data.
problem Insufficient evaluation of change point detection algorithms on real-world time series.
method Developed a data set of 37 time series from various domains, annotated by human experts, and evaluated 14 algorithms using consistency metrics.
result Demonstrates the need for better evaluation methods in change point detection.
This research integrates human interaction into reinforcement learning to improve sample efficiency and real-time learning.
problem Current reinforcement learning requires thousands of samples to converge, and is prone to catastrophic failures.
method Integrates human interaction modalities (demonstrations, interventions, evaluations) into the reinforcement learning loop.
result Human interaction accelerates learning and improves sample efficiency.
Improves AI system's understanding of human inputs by creating better examples.
problem AI systems misinterpret human inputs, leading to inefficiencies.
method Developed a conditional convolutional autoencoder (CCAE) to generate better examples.
result Generated examples lead to lower error rates and require less effort to create.
Online symptom checkers have significant potential to improve patient care, however their reliability and accuracy remain variable. We hypothesised that an artificial intelligence (AI) powered triage and diagnostic system would compare favourably with human doctors with respect to triage and diagnostic accuracy. We per…
Transformer models improve financial sentiment measurement.
problem Capturing nuanced sentiment from financial news articles.
method Transformer-based language models for sentiment classification and aggregation.
result Transformer models outperform traditional dictionary-based methods in sentiment classification.
Autonomous systems can substantially enhance a human's efficiency and effectiveness in complex environments. Machines, however, are often unable to observe the preferences of the humans that they serve. Despite the fact that the human's and machine's objectives are aligned, asymmetric information, along with heterogene…
In this paper, we propose a framework that enables a human teacher to shape a robot behaviour by interactively providing it with unlabeled instructions. We ground the meaning of instruction signals in the task-learning process, and use them simultaneously for guiding the latter. We implement our framework as a modular …