Limits...
Implementation of olfactory bulb glomerular-layer computations in a digital neurosynaptic core.

Imam N, Cleland TA, Manohar R, Merolla PA, Arthur JV, Akopyan F, Modha DS - Front Neurosci (2012)

Bottom Line: Our system is based on a digital neuromorphic chip consisting of 256 leaky-integrate-and-fire neurons, 1024 × 256 crossbar synapses, and address-event representation communication circuits.The neural circuits configured in the chip reflect established connections among mitral cells, periglomerular cells, external tufted cells, and superficial short-axon cells within the olfactory bulb, and accept input from convergent sets of sensors configured as olfactory sensory neurons.Our circuits, consuming only 45 pJ of active power per spike with a power supply of 0.85 V, can be used as the first stage of processing in low-power artificial chemical sensing devices inspired by natural olfactory systems.

View Article: PubMed Central - PubMed

Affiliation: Computer Systems Lab, Department of Electrical and Computer Engineering, Cornell University Ithaca, NY, USA.

ABSTRACT
We present a biomimetic system that captures essential functional properties of the glomerular layer of the mammalian olfactory bulb, specifically including its capacity to decorrelate similar odor representations without foreknowledge of the statistical distributions of analyte features. Our system is based on a digital neuromorphic chip consisting of 256 leaky-integrate-and-fire neurons, 1024 × 256 crossbar synapses, and address-event representation communication circuits. The neural circuits configured in the chip reflect established connections among mitral cells, periglomerular cells, external tufted cells, and superficial short-axon cells within the olfactory bulb, and accept input from convergent sets of sensors configured as olfactory sensory neurons. This configuration generates functional transformations comparable to those observed in the glomerular layer of the mammalian olfactory bulb. Our circuits, consuming only 45 pJ of active power per spike with a power supply of 0.85 V, can be used as the first stage of processing in low-power artificial chemical sensing devices inspired by natural olfactory systems.

No MeSH data available.


Related in: MedlinePlus

The contrast enhancement function in mitral cells on the chip in response to sensor inputs. Mitral cells corresponding to moderately activated sensors are predominantly inhibited (reducing the spike rate of some cells to nearly 20% below baseline activity) by feed-forward PGo inhibition (moderate activation), whereas mitral cells excited by strongly activated sensors overcome the saturated PGo inhibition and exhibit net excitation (strong activation). The leak portion of the “inhibition + leak” curve represents charge leakage out of the cell between successive excitatory spikes received from OSN inputs, and therefore is inversely related to the activity of the OSNs. For strongly activated sensors, the excitation delivered to the mitral cells is of sufficiently high frequency to overcome the combined effects of PGo inhibition and the leak current. Figure adapted from Cleland and Sethupathy (2006).
© Copyright Policy - open-access
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC3368244&req=5

Figure 9: The contrast enhancement function in mitral cells on the chip in response to sensor inputs. Mitral cells corresponding to moderately activated sensors are predominantly inhibited (reducing the spike rate of some cells to nearly 20% below baseline activity) by feed-forward PGo inhibition (moderate activation), whereas mitral cells excited by strongly activated sensors overcome the saturated PGo inhibition and exhibit net excitation (strong activation). The leak portion of the “inhibition + leak” curve represents charge leakage out of the cell between successive excitatory spikes received from OSN inputs, and therefore is inversely related to the activity of the OSNs. For strongly activated sensors, the excitation delivered to the mitral cells is of sufficiently high frequency to overcome the combined effects of PGo inhibition and the leak current. Figure adapted from Cleland and Sethupathy (2006).

Mentions: On-center/inhibitory surround contrast enhancement has been observed in mitral cells (second-order principal neurons; Figure 1) along a trajectory of odor similarity (Yokoi et al., 1995). Critically, such maps of similarity among the aggregate chemoreceptive fields of all glomeruli in the olfactory bulb are intrinsically high-dimensional. That is, if OR chemoreceptive fields are compactly modeled (e.g., as hyperellipses), there are no low-dimensional solutions to the problem of mapping all of these fields into a common space such that all of their similarity relationships (degrees of pairwise overlap) are respected. Consequently, the glomerular layer does not contain a true map of odor similarity or perceptual quality, but does comprise a lookup table of OR chemoreceptive fields (Murthy, 2011). This becomes important because decorrelation operations such as contrast enhancement are similarity-dependent computations; that is, to sharpen an odor representation, the contrast enhancement operation must operate within this high-dimensional similarity space such that the “edges” of the odor representation are selectively inhibited. Two-dimensional algorithms such as lateral inhibition are wholly ineffective in this environment (Cleland and Sethupathy, 2006), except to the extent that disordered inhibition (which does not generate an inhibitory surround) can decorrelate such representations by broadly reducing sensitivity (Cleland and Linster, 2012). In the olfactory bulb, high-dimensional contrast enhancement arises via an intra-glomerular, non-topographical computation based on OSN input to mitral cells paired with feed-forward inhibition via olfactory nerve-driven periglomerular (PGo) cells. Briefly, at lower relative input levels (i.e., compared to the overall activity across all glomeruli), the feed-forward inhibition dominates, inhibiting mitral cells below baseline to generate the inhibitory surround. At higher relative input levels, direct OSN excitation overcomes PGo-mediated inhibition such that the mitral cell is excited by the odor. The net effect is that the most highly excited mitral cells are activated, whereas mitral cells that receive only moderate levels of OSN input are selectively and disproportionately inhibited, thereby generating surround inhibition in the native similarity space of the sensor array. Importantly, from an artificial systems perspective, this property enables contrast enhancement to be performed on input from any arbitrary sensor complement without the need for sensor-specific programming. The chip described herein replicates this functionality (Figure 9), while replacing the shunting inhibition-based biological mechanism with a low-power spike-based algorithm.


Implementation of olfactory bulb glomerular-layer computations in a digital neurosynaptic core.

Imam N, Cleland TA, Manohar R, Merolla PA, Arthur JV, Akopyan F, Modha DS - Front Neurosci (2012)

The contrast enhancement function in mitral cells on the chip in response to sensor inputs. Mitral cells corresponding to moderately activated sensors are predominantly inhibited (reducing the spike rate of some cells to nearly 20% below baseline activity) by feed-forward PGo inhibition (moderate activation), whereas mitral cells excited by strongly activated sensors overcome the saturated PGo inhibition and exhibit net excitation (strong activation). The leak portion of the “inhibition + leak” curve represents charge leakage out of the cell between successive excitatory spikes received from OSN inputs, and therefore is inversely related to the activity of the OSNs. For strongly activated sensors, the excitation delivered to the mitral cells is of sufficiently high frequency to overcome the combined effects of PGo inhibition and the leak current. Figure adapted from Cleland and Sethupathy (2006).
© Copyright Policy - open-access
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC3368244&req=5

Figure 9: The contrast enhancement function in mitral cells on the chip in response to sensor inputs. Mitral cells corresponding to moderately activated sensors are predominantly inhibited (reducing the spike rate of some cells to nearly 20% below baseline activity) by feed-forward PGo inhibition (moderate activation), whereas mitral cells excited by strongly activated sensors overcome the saturated PGo inhibition and exhibit net excitation (strong activation). The leak portion of the “inhibition + leak” curve represents charge leakage out of the cell between successive excitatory spikes received from OSN inputs, and therefore is inversely related to the activity of the OSNs. For strongly activated sensors, the excitation delivered to the mitral cells is of sufficiently high frequency to overcome the combined effects of PGo inhibition and the leak current. Figure adapted from Cleland and Sethupathy (2006).
Mentions: On-center/inhibitory surround contrast enhancement has been observed in mitral cells (second-order principal neurons; Figure 1) along a trajectory of odor similarity (Yokoi et al., 1995). Critically, such maps of similarity among the aggregate chemoreceptive fields of all glomeruli in the olfactory bulb are intrinsically high-dimensional. That is, if OR chemoreceptive fields are compactly modeled (e.g., as hyperellipses), there are no low-dimensional solutions to the problem of mapping all of these fields into a common space such that all of their similarity relationships (degrees of pairwise overlap) are respected. Consequently, the glomerular layer does not contain a true map of odor similarity or perceptual quality, but does comprise a lookup table of OR chemoreceptive fields (Murthy, 2011). This becomes important because decorrelation operations such as contrast enhancement are similarity-dependent computations; that is, to sharpen an odor representation, the contrast enhancement operation must operate within this high-dimensional similarity space such that the “edges” of the odor representation are selectively inhibited. Two-dimensional algorithms such as lateral inhibition are wholly ineffective in this environment (Cleland and Sethupathy, 2006), except to the extent that disordered inhibition (which does not generate an inhibitory surround) can decorrelate such representations by broadly reducing sensitivity (Cleland and Linster, 2012). In the olfactory bulb, high-dimensional contrast enhancement arises via an intra-glomerular, non-topographical computation based on OSN input to mitral cells paired with feed-forward inhibition via olfactory nerve-driven periglomerular (PGo) cells. Briefly, at lower relative input levels (i.e., compared to the overall activity across all glomeruli), the feed-forward inhibition dominates, inhibiting mitral cells below baseline to generate the inhibitory surround. At higher relative input levels, direct OSN excitation overcomes PGo-mediated inhibition such that the mitral cell is excited by the odor. The net effect is that the most highly excited mitral cells are activated, whereas mitral cells that receive only moderate levels of OSN input are selectively and disproportionately inhibited, thereby generating surround inhibition in the native similarity space of the sensor array. Importantly, from an artificial systems perspective, this property enables contrast enhancement to be performed on input from any arbitrary sensor complement without the need for sensor-specific programming. The chip described herein replicates this functionality (Figure 9), while replacing the shunting inhibition-based biological mechanism with a low-power spike-based algorithm.

Bottom Line: Our system is based on a digital neuromorphic chip consisting of 256 leaky-integrate-and-fire neurons, 1024 × 256 crossbar synapses, and address-event representation communication circuits.The neural circuits configured in the chip reflect established connections among mitral cells, periglomerular cells, external tufted cells, and superficial short-axon cells within the olfactory bulb, and accept input from convergent sets of sensors configured as olfactory sensory neurons.Our circuits, consuming only 45 pJ of active power per spike with a power supply of 0.85 V, can be used as the first stage of processing in low-power artificial chemical sensing devices inspired by natural olfactory systems.

View Article: PubMed Central - PubMed

Affiliation: Computer Systems Lab, Department of Electrical and Computer Engineering, Cornell University Ithaca, NY, USA.

ABSTRACT
We present a biomimetic system that captures essential functional properties of the glomerular layer of the mammalian olfactory bulb, specifically including its capacity to decorrelate similar odor representations without foreknowledge of the statistical distributions of analyte features. Our system is based on a digital neuromorphic chip consisting of 256 leaky-integrate-and-fire neurons, 1024 × 256 crossbar synapses, and address-event representation communication circuits. The neural circuits configured in the chip reflect established connections among mitral cells, periglomerular cells, external tufted cells, and superficial short-axon cells within the olfactory bulb, and accept input from convergent sets of sensors configured as olfactory sensory neurons. This configuration generates functional transformations comparable to those observed in the glomerular layer of the mammalian olfactory bulb. Our circuits, consuming only 45 pJ of active power per spike with a power supply of 0.85 V, can be used as the first stage of processing in low-power artificial chemical sensing devices inspired by natural olfactory systems.

No MeSH data available.


Related in: MedlinePlus