1: \begin{abstract}

2: Pool-based active learning promises more efficient use of noisy oracle responses by querying informative data points. 

3: In this paper, we build a connection between active sequential hypothesis testing, a framework initiated by \cite{chernoff1959sequential}, and active regression. 

4: For active testing, using techniques introduced by \cite{naghshvar2013active}, we obtain a novel sample complexity bound for Chernoff’s original procedure, uncovering non-asymptotic terms that reduce in significance as the allowed error probability $\delta \rightarrow 0$. Initially proposed for testing among finitely many hypotheses, we obtain the analogue of Chernoff sampling for the case when the hypotheses belong to a compact space. This allows us to use it in estimating the parameter in active regression. 

5: Using techniques introduced by \cite{chaudhuri2015convergence}, we show that the risk of our procedure converges to the optimal value. 

6: %Our risk bound gives another interpretation for Chernoff sampling--- it reduces the loss suffered if the procedure was interrupted at any time by an adversary who then tests the estimate on the worst-possible subset of data points in the pool. 

7: Unlike existing methods that typically change the sampling proportion in stages, our procedure is fully adaptive. We demonstrate its potential by actively learning neural network models and in real-data linear and non-linear regression problems, where our general-purpose approach outperforms state-of-the-art methods.

8: \end{abstract}

9: