Task-oriented dialog systems need to know when a query falls outside their range of supported intents, but current text classification corpora only define label sets that cover every example. We introduce a new dataset that includes queries that are out-of-scopei.e., queries that do not fall into any of the system's supported intents. This poses a new challenge because models cannot assume that every query at inference time belongs to a system-supported intent class. Our dataset also covers 150 intent classes over 10 domains, capturing the breadth that a production taskoriented agent must handle. We evaluate a range of benchmark classifiers on our dataset along with several different out-of-scope identification schemes. We find that while the classifiers perform well on in-scope intent classification, they struggle to identify out-of-scope queries. Our dataset and evaluation fill an important gap in the field, offering a way of more rigorously and realistically benchmarking text classification in task-driven dialog systems.

In this paper, we describe the design and testing of a system for recording electroneurographic signals (ENG) from a multielectrode nerve cuff (MEC). This device, which is an extension of the conventional nerve signal recording cuff, enables ENG to be classified by action potential velocity. In addition to electrical measurements, we provide preliminary in vitro data obtained from frogs that demonstrate the validity of the technique for the first time. Since typical ENG signals are extremely small, on the order of 1 1 microV, very low-noise, high-gain amplifiers are required. The ten-channel system we describe was realized in a 0.8 microm CMOS technology and detailed measured results are presented. The overall gain is 10 000 and the total input-referred root mean square (rms) noise in a bandwidth 1 Hz-5 kHZ is 291 nV. The active area is 12 mm(2) and the power consumption is 24 mW from +/-2.5 V power supplies.

This paper describes the improvements to the theory of velocity selective recording and some simulation results. In this method, activity in different groups of axons is discriminated by their propagation velocity. A multi-electrode cuff and an array of amplifiers produce multiple neural signals; if artificial delays are inserted and the signals are added, the activity in axons of the matched velocity are emphasized. We call this intrinsic velocity selective recording. However, simulation shows that interpreting the time signals is then not straight-forward and the selectivity Q(v) is low. New theory shows that bandpass filters improve the selectivity and explains why this is true in the time domain. A simulation study investigates the limits on the available velocity selectivity both with and without additive noise and with reasonable sampling rates and analogue-to-digital conversion parameters. Bandpass filters can improve the selectivity by factors up to 7 but this depends on the speed of the action potential and the signal-to-noise ratio.

No abstract

Figure 1: BimodalGaze enables users to point by gaze and to seamlessly refine the cursor position with head movement. A: In Gaze Mode, the cursor (yellow) follows where the user looks but may not be sufficiently accurate. B: The pointer automatically switches into Head Mode (green) when gestural head movement is detected. C: The pointer automatically switches back into Gaze Mode when the user redirects their attention. Note that the Head Mode is only invoked when needed for adjustment of the cursor. Any natural head movement associated with a gaze shift is filtered and does not cause a mode switch.