Approximate Function Evaluation via Multi-Armed Bandits

03/18/2022
by   Tavor Z. Baharav, et al.
8

We study the problem of estimating the value of a known smooth function f at an unknown point μ∈ℝ^n, where each component μ_i can be sampled via a noisy oracle. Sampling more frequently components of μ corresponding to directions of the function with larger directional derivatives is more sample-efficient. However, as μ is unknown, the optimal sampling frequencies are also unknown. We design an instance-adaptive algorithm that learns to sample according to the importance of each coordinate, and with probability at least 1-δ returns an ϵ accurate estimate of f(μ). We generalize our algorithm to adapt to heteroskedastic noise, and prove asymptotic optimality when f is linear. We corroborate our theoretical results with numerical experiments, showing the dramatic gains afforded by adaptivity.

READ FULL TEXT

Please sign up or login with your details

Forgot password? Click here to reset