Robots learn conservatively from human corrections, avoiding unintended changes to their objectives.
problem Robots learn from human corrections but may not align with the intended objectives due to misspecified hypothesis spaces.
method Robots reason in real-time about the relevance of human corrections to their hypothesis space, learning more conservatively.
result Robots can avoid unintended learning from human corrections, improving alignment with intended objectives.
Paper addresses hypothesis space misspecification in learning from human demonstrations and corrections.
problem Hypothesis space misspecification in learning from human demonstrations and corrections.
method Reason explicitly about how well the robot can explain human inputs given its hypothesis space.
result Demonstrates method on a 7 DOF robot manipulator.
Proposes PA-DSL for correcting noisy human labels in automated data labeling.
problem Noisy human labels in automated data labeling.
method Uses adjudicated cases to correct noisy human labels and debias analyses.
result Maintains nominal coverage and reduces RMSE by 10-17% relative to using only adjudicated labels.
New algorithm learns Gaussian policies from corrective human feedback, outperforming current methods.
problem Learning from corrective human feedback for complex systems.
method Gaussian Process Coach (GPC) that uses Gaussian Processes and policy uncertainty for optimal feedback selection and learning rate adaptation.
result Demonstrated superior performance in OpenAI Gym benchmarks compared to COACH.
Deep Reinforcement Learning (DRL) has become a powerful strategy to solve complex decision making problems based on Deep Neural Networks (DNNs). However, it is highly data demanding, so unfeasible in physical systems for most applications. In this work, we approach an alternative Interactive Machine Learning (IML) stra…
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.
Theoretical analysis shows LLMs can self-correct responses through in-context learning.
problem Understanding how large language models improve through self-correction.
method Theoretical analysis based on simplified alignment task, focusing on softmax attention, multi-head attention, and MLP blocks.
result LLMs can refine responses in an in-context way when given accurate self-examinations as rewards.
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.
Paper proposes an efficient method for bounding box annotation in object detection.
problem Manual annotation of bounding boxes is tedious and resource-intensive.
method Iterative training of object detector on small batches of labeled images, with human annotator correcting errors.
result Significant reduction in human annotation effort, up to 75%.
Optimal allocation of human effort to correct AI assessments in decision-making.
problem How to allocate costly human effort to correct noisy or biased AI-generated assessments.
method Decision-theoretic framework treating AI assessments as signals and human judgments as costly information. Developed estimation procedures under nonparametric and linear models.
result Our approach substantially outperforms LLM-only predictions and achieves performance comparable to full human review while using only 20-30% of the human information.
This work advances collaborative decision making by combining human and AI strengths in uncertainty quantification.
problem Current AI lacks robust decision-making capabilities under uncertainty, especially in high-stakes contexts.
method Introduces Human AI Collaborative Uncertainty Quantification (HACUQ) framework, formalizing AI-human collaboration and developing calibration algorithms.
result Optimal collaborative prediction sets follow a two-threshold structure, and online adaptation algorithms can adapt to evolving human behavior.
Explainable AI improves human decision accuracy but does not enhance it significantly.
problem Improving human decision-making through explainable AI.
method Comparing human decision accuracy with and without AI predictions, including or excluding explanations.
result Providing AI predictions improves human decision accuracy, but explanations do not significantly enhance it.
Develops methods to correct bias in AI feedback for more accurate alignment.
problem Systematic bias in AI feedback compared to human labels.
method Two debiased alignment methods: DDPO and DIPO.
result Methods improve alignment efficiency and performance close to human-labeled data.
Paper proposes method to evaluate AI in ranking tasks with individual differences.
problem Difficulty in evaluating AI in tasks where correct answers vary by individual.
method Probabilistic model of human ranking behavior and efficient computation method.
result Demonstrates AI ranking results can be distinguished from human-generated ones.
Paper stabilizes generative model training with synthetic data.
problem Self-consuming loops in generative model training.
method Introducing an idealized correction function and self-correction functions.
result Self-consuming loops can be exponentially more stable with the right correction.
Develops a new tensor model for clustering with degree correction.
problem Clustering with unknown degree heterogeneity in multiway data.
method Degree-corrected tensor block model with estimation guarantees.
result Demonstrates an intrinsic statistical-to-computational gap for tensors of order three or greater.
New method uses explicit human demonstrations to teach missing features in reward learning.
problem Reward learning methods rely on handcrafted features, limiting their ability to adapt to new or unexplained corrections.
method Introduces human input guiding the robot from states with missing features to states without, teaching the feature explicitly and integrating it into the reward function.
result Decreases sample complexity and improves generalization of the learned reward over deep IRL baseline.
We present an approach to interactive-predictive neural machine translation that attempts to reduce human effort from three directions: Firstly, instead of requiring humans to select, correct, or delete segments, we employ the idea of learning from human reinforcements in form of judgments on the quality of partial tra…
A new L2D system produces calibrated probabilities of expert correctness without sacrificing accuracy.
problem Calibration of learning to defer systems for safety.
method One-vs-all classifiers with a consistent surrogate loss function.
result Proposes a calibrated L2D system that outperforms existing methods in accuracy and calibration.
New approach shows AI can adapt like toddlers by correcting old knowledge.
problem Lack of adaptability in AI models compared to humans and animals.
method Casting adaptation as posterior correction and using Bayesian Learning Rule.
result AI can learn to adapt quickly by using posterior correction.
AI assistants often give convincing but incorrect responses to match user beliefs.
problem Sycophancy in AI assistants that use human feedback.
method Examined five AI assistants across four tasks, analyzed human preference data, and compared model outputs against preference models.
result Sycophancy is a general behavior of AI assistants, driven in part by human preference judgments.
The concept of progress has characterized human society from millennia. However, this concept is elusive and too often given for certain. The goal of this paper is to suggest a general definition of human progress that satisfies, whenever possible the conditions of independence, generality, epistemological applicabilit…
Study improves understanding of what makes machine learning explanations human-interpretable.
problem Understanding what makes explanations human-interpretable in machine learning systems.
method Controlled human-subject experiments to identify regularizers for interpretability across three tasks.
result Cognitive chunks affect performance more than variable repetitions, suggesting common design principles.
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.
This paper improves deep learning model consistency through ensemble methods.
problem Consistency and correct-consistency issues in deep learning models.
method Formal definition of consistency and correct-consistency, proving ensemble improvement, proposing dynamic snapshot ensemble method.
result Ensemble methods can improve correct-consistency of deep learning models.
Study evaluates human vs. machine review generation, finds human assessments correlate better with lexical overlaps.
problem Evaluating natural language generation models for online reviews is challenging and inconsistent.
method Compared human evaluators with various automated evaluation methods, including discriminative and word overlap metrics.
result Human evaluators do not correlate well with discriminative evaluators, but correlate better with lexical overlaps.
We consider the problem of learning object arrangements in a 3D scene. The key idea here is to learn how objects relate to human poses based on their affordances, ease of use and reachability. In contrast to modeling object-object relationships, modeling human-object relationships scales linearly in the number of objec…
New method uses feedback to improve deep reinforcement learning efficiency.
problem Vast amounts of data needed for reasonable performance in deep reinforcement learning.
method Binary corrective feedback combined with probabilistic conditional exploration.
result Achieves drastic improvements in sample efficiency and robustness.
Game theory improves smart road sign security against small perturbations.
problem Ensuring smart road signs are secure from small-scale adversarial attacks.
method Integrates game theory into smart road sign classification to detect imperceptible perturbations.
result Proposes a randomized detection strategy to ensure robustness against worst-case attackers.
While linear mixed model (LMM) has shown a competitive performance in correcting spurious associations raised by population stratification, family structures, and cryptic relatedness, more challenges are still to be addressed regarding the complex structure of genotypic and phenotypic data. For example, geneticists hav…
IAL uses interactive learning to improve model performance with minimal human feedback.
problem Overfitting and high human interaction cost in training neural networks.
method IAL framework with NAP attention generator and reranking algorithm.
result IAL significantly outperforms baselines with less retraining and human interaction.
Aerial robot estimates human pose and path using dynamic classifier selection.
problem Estimating human pose and trajectory from aerial video.
method Dynamic classifier selection architecture; perspective correction; HOG and CNN features; 64 pose-viewpoint classes.
result Dynamic classifier selection improves efficiency and accuracy.
TIMELY improves consistency in labeling blood cell images.
problem Inconsistent labeling of blood cells in microscopy images leads to unreliable diagnoses.
method TIMELY combines pseudotime inference and hidden Markov trees to correct labeling mistakes.
result TIMELY outperforms baseline methods in identifying and correcting inconsistent labels.
A new watermarking method corrects bias in language models using maximal coupling.
problem Correcting bias in language model token distributions.
method Maximal coupling to balance bias correction and text quality.
result Outperforms prior techniques in preserving text quality and detectability.
STAS selects optimal spatio-temporal scales for bias correction in precipitation forecasts.
problem Limited prior data and fixed ST scale in existing BCoPs lead to biases in numerical weather predictions.
method End-to-end deep-learning BCoP model STAS with SFM/TFM to automatically adjust spatial and temporal scales.
result STAS outperforms 8 published BCoP methods on threat scores (TS).
In this study, we propose a novel deep neural network and its supervised learning method that uses a feedforward supervisory signal. The method is inspired by the human visual system and performs human-like association-based learning without any backward error propagation. The feedforward supervisory signal that produc…
Machines, not humans, are the world's dominant knowledge accumulators but humans remain the dominant decision makers. Interpreting and disseminating the knowledge accumulated by machines requires expertise, time, and is prone to failure. The problem of how best to convey accumulated knowledge from computers to humans i…
DAL uses disentanglement for automatic labeling in GAN-based active learning.
problem Reducing human labeling in GAN-based active learning.
method DAL leverages disentanglement in InfoGAN to automatically label datapoints, deciding human labeling based on disagreement with InfoGAN labels and label correction.
result DAL achieves better performance than existing GAN-based active learning approaches on image classification tasks.
A method for training autonomous vehicles using continuous human feedback to avoid sub-optimal decisions.
problem Training autonomous vehicles with limited and potentially sub-optimal human demonstrations.
method Continuous scalar feedback for each action to learn from sub-optimal demonstrations and evaluative feedback.
result The proposed method outperforms supervised learning on positive examples alone and learns from sub-optimal demonstrations.
SRPO improves AI alignment with human preferences through self-improvement and task-independent optimization.
problem AI models trained with RLHF lack self-correction mechanisms and struggle with task generalization.
method SRPO formulates the preference learning problem as a min-max objective, optimizing a self-improvement policy and a generative policy in an adversarial fashion, making the solution task-independent.
result SRPO outperforms existing methods, achieving 90% AI Win-Rate on XSum and 56% on Arena-Hard prompts after a single revision.
This research explores inductive biases for deep learning to improve AI's higher-level cognition.
problem Current AI struggles with flexible out-of-distribution and systematic generalization.
method Examines and proposes new inductive biases for deep learning.
result Identifies specific inductive biases for higher-level sequential processing.
k-Rater reliability corrects under-reporting of aggregated data reliability.
problem Under-reporting of data reliability in aggregated ratings.
method k-Rater reliability (kRR) as a multi-rater generalization of IRR.
result kRR provides a more accurate measure of reliability for aggregated datasets.
GICDM corrects hubness in embedding spaces for better generative model evaluation.
problem Hubness phenomenon distorts distances in high-dimensional embedding spaces.
method Generative ICDM (GICDM) using multi-scale extension to correct neighborhood estimation.
result GICDM resolves hubness-induced failures and improves metric behavior.
We proposed a probabilistic approach to joint modeling of participants' reliability and humans' regularity in crowdsourced affective studies. Reliability measures how likely a subject will respond to a question seriously; and regularity measures how often a human will agree with other seriously-entered responses coming…
XAI methods fail to explain ML models reliably.
problem Current XAI methods fail to provide reliable explanations for ML models.
method Formally define problems and design methods accordingly.
result Diverse notions of explanation correctness and metrics needed.
The paper proposes a method to evaluate superhuman models by checking for logical inconsistencies.
problem Evaluating superhuman models when ground truth is hard to obtain.
method A framework using consistency checks to identify logical inconsistencies in model decisions.
result Logical inconsistencies can be discovered in superhuman model decisions across various tasks.
Method curates cost-effective, high-quality datasets using AI models.
problem Costly manual labeling of datasets.
method Probably Approximately Correct Labels (PACL) method.
result Curates high-quality datasets with low overall labeling error.
Background: A universal unanswered question in neuroscience and machine learning is whether computers can decode the patterns of the human brain. Multi-Voxels Pattern Analysis (MVPA) is a critical tool for addressing this question. However, there are two challenges in the previous MVPA methods, which include decreasing…