The Human Kernel

By Andrew Gordon Wilson, Christoph Dann, Christopher G. Lucas, Eric P. Xing. 2015

69 citations recorded by the source corpus.

Bayesian nonparametric models, such as Gaussian processes, provide a compelling framework for automatic statistical modelling: these models have a high degree of flexibility, and automatically calibrated complexity. However, automating human expertise remains elusive; for example, Gaussian processes with standard kernels struggle on function extrapolation problems that are trivial for human learners. In this paper, we create function extrapolation problems and acquire human responses, and then design a kernel learning framework to reverse engineer the inductive biases of human learners across a set of behavioral experiments. We use the learned kernels to gain psychological insights and to extrapolate in human-like ways that go beyond traditional stationary and polynomial kernels. Finally, we investigate Occam's razor in human and Gaussian process based function learning.

Subfields: Artificial Intelligence

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

Citations in the corpus

Interactive machine learning for health informatics: when do we need the human-in-the-loop? (2016, background): “Consequently, iML-approaches, by integrating a human-into-the-loop (e.g., a human kernel [24], thereby making use of human cognitive abilities, seem to be a promising approach. iML-approaches can be of particular interest to solve problems in health informatics, where we are lacking big data sets,…”
From Anecdotal Evidence to Quantitative Evaluation Methods: A Systematic Review on Evaluating Explainable AI (2022, background): “[279] found that humans favor smooth and simpler functions, and have inductive biases towards recognizable patterns, such as step functions or a sawtooth pattern.”
A glass-box interactive machine learning approach for solving NP-hard problems with the human-in-the-loop (2017, methodology): “By integrating a human-in-the-loop (e.g., a human kernel (Wilson et al., 2015)), or the involvement of a human directly into the algorithm, iML is deﬁned by ”algorithms which interact with agents and can optimize their learning behaviour through this interaction – where the agents can be humans…”
Introduction to MAchine Learning & Knowledge Extraction (MAKE) (2017, background): “, a human kernel [33]), or the involvement of a human directly into the machine-learning algorithm [39], thereby making use of human cognitive abilities, is a promising approach.”
Towards interactive Machine Learning (iML): Applying Ant Colony Algorithms to Solve the Traveling Salesman Problem with the Human-in-the-Loop Approach (2016, unknown): “An interesting experiment was performed by [14]: humans were presented functions drawn from Gaussian Processes (GP) with known kernels in sequence and asked to make extrapolations.”
A Roadmap for a Rigorous Science of Interpretability (2017, background): “Are some functions more natural to humans than others [Wilson et al., 2015, Schulz et al., 2016]?”
Towards A Rigorous Science of Interpretable Machine Learning (2017, background): “Are some functions more natural to humans than others [Wilson et al., 2015, Schulz et al., 2016]?”
Human-in-the-Loop Interpretability Prior (2018, background): “Examples include kernel learning [28, 31], preference based reinforcement learning [32, 5] and human based genetic algorithms [12].”
A tutorial on Gaussian process regression: Modelling, exploring, and exploiting functions (2016, methodology): “There are recent efforts to make hyper-parameter estimation fully Bayesian, for example by using Stan (Flaxman, Gelman, Neill, Smola, Vehtari, and Wilson, 2015), which are promising to result in more robust estimates by additionally providing uncertainty estimates for the obtained parameters.”
On the Origin of Deep Learning (2017, background): “Some established work to modeling human factors (Wilson et al., 2015) could be helpful.”