NEWS & VIEWS RESEARCH are generated by multiple thalamic inputs neurons across the visual–cortical map . that have temporally different responses to Whatever the answers are, it is becoming the stimulus (Fig. 1). Thalamic inputs that increasingly clear that the visual cortex gen- respond slowly to visual stimuli generate slow erates stimulus selectivity, such as prefer- responses in cortical regions, whereas those ences for direction and orientation, through responding faster generate fast responses. thalamo–cortical convergence. Lien and Lien and Scanziani’s results, taken together Scanziani’s work shows that this mechanism 3–10 with previous work , raise the interesting pos- is better preserved across mammals than was sibility that cortical direction selectivity is gen- previously thought. ■ 50 Years Ago erated through a common mechanism — the convergence of temporally diverse thalamic Jose Manuel Alonso is in the College of Reading aids for the blind have inputs — in rodents, cats and primates. But as Optometry, State University of New York, so far involved the use of intact with all research, some questions remain open. New York, New York 10036, USA. sensory pathways and have For instance, the authors focus their study em ail: email@example.com progressed little beyond Braille on the middle layers of the visual cortex, and tape-recorded “talking-books”. 1. Hubel, D. H. & Wiesel, T. N. J. Physiol. 160, 106–154 which receive the bulk of the thalamic input . Both these systems are quite (1962). As Lien and Scanziani show, many thalamic expensive … and both are slow in 2. Lien, A. D. & Scanziani, M. Nature 558, 80–86 inputs in these middle cortical layers are not (2018). terms of information transfer to the 3. Alonso, J.-M., Usrey, W. M. & Reid, R. C. J. Neurosci. direction selective, but their combined activity reader … At a recent meeting of the 21, 4002–4015 (2001). is. It remains unclear whether thalamic inputs Physiological Society, Brindley and 4. Ferster, D., Chung, S. & Wheat, H. Nature 380, that target other cortical layers (or serve other Lewin demonstrated a device for 249–252 (1996). 5. Saul, A. B. & Humphrey, A. L. J. Neurophysiol. 64, functions) can encode direction selectivity stimulating the visual cortex of man 206–224 (1990). through different mechanisms. For example, directly … Essentially it consists 6. Saul, A. B. & Humphrey, A. L. J. Neurophysiol. 68, neurons in the superficial layers of the cor- of an array of radio receivers, 1190–1208 (1992). tex might derive their direction selectivity 7. Stanley, G. B. et al. J. Neurosci. 32, 9073–9088 encapsulated in silicone rubber and (2012). from thalamic neurons that are themselves screwed to the skull … Activation 8. Reid, R. C., Soodak, R. E. & Shapley, R. M. direction selective . of a receiver stimulated the cortex: J. Neurophysiol. 66, 505–529 (1991). It is also known that thalamic inputs to the transmission was in the form of 9. Livingstone, M. S. Neuron 20, 509–526 (1998). 10. McLean, J. & Palmer, L. A. Vision Res. 29, 675–679 visual cortex are arranged by their receptive- a train of short (200 μs) pulses … (1989). field position — inputs that have receptive fields it does at least seem feasible to 11. Lorente de No, R. In Physiology of the Nervous close to one another in the field of view are clus- transmit visual information directly System (ed. Fulton, J.) 291–340 (Oxford Univ. Press, tered together. However, it is not yet known to the central visual pathways of the 1938). 12. Cruz-Martín, A. et al. Nature 507, 358–361 (2014). whether the thalamic inputs are also arranged recently blind. 13. Kremkow, J., Jin, J., Wang, Y. & Alonso, J. M. Nature according to their temporal properties. If so, this From Nature 8 June 1968 533, 52–57 (2016). could explain why spatial position and direction preference tend to change together in different This article was published online on 23 May 2018. 100 Years Ago ENGINEERING It happened last week that about 1 lb. of fresh lamb was put into an oven at night in order that it Two artificial synapses might be cooked by morning on the “hay-box” principle. It was in a casserole, with a little water. are better than one Similar treatment in the same oven on previous occasions had been Emerging nanoelectronic devices could revolutionize artificial neural networks, but very successful. At about 5 a.m. their hardware implementations lag behind those of their software counterparts. An the casserole was examined, and approach has been developed that tips the scales in their favour. See Article p .60 the broth was found to be very well tasted, and the whole smelt fresh and good, but the meat when tested GINA C. AD AM can perform as well as can software-based with a fork was not tender, and networks running on ordinary computers, the fat (of which there was a good nspired by the brain’s neural networks, while consuming much less power. deal) was entirely unmelted. The scientists have for decades tried to Artificial neural networks are not casserole was returned to the oven Iconstruct electronic circuits that can programmed in the same way as conventional (then quite cool) and taken out process large amounts of data. However, it computers. Just as humans learn from experi- again after breakfast. The contents has been difficult to achieve energy-efficient ence, these networks acquire their functions were then found to be smelling implementations of artificial neurons and from data obtained during a training process. most offensively, as if extremely synapses (connections between neurons). Image classification, which involves learning “high”. The fat was melted. The On page 60, Ambrogio et al. report an arti- and memory, requires thousands of artificial meat and broth were judged quite ficial neural network containing more than synapses. The states (electrical properties) of unfit for human food. I wonder if 200,000 synapses that can classify complex these synapses need to be programmed quickly any of your readers would explain collections of images. The authors’ work dem- and then retained for future network operation. this curious development. onstrates that hardware-based neural networks Nanoscale synaptic devices that have From Nature 6 June 1918 that use emerging nanoelectronic devices programmable electrical resistance, such 7 J U N E 2018 | V O L 558 | N A T U R E | 39 © 201 8 Mac m ill an Publi shers Li m it ed, part of Spri nger Nat ur e. A ll ri ghts r eser ved. © 201 8 Mac m ill an Publi shers Li m it ed, part of Spri nger Nat ur e. A ll ri ghts r eser ved.
Nature – Springer Journals
Published: Jun 5, 2018
It’s your single place to instantly
discover and read the research
that matters to you.
Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.
All for just $49/month
Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly
Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.
Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.
Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.
All the latest content is available, no embargo periods.
“Hi guys, I cannot tell you how much I love this resource. Incredible. I really believe you've hit the nail on the head with this site in regards to solving the research-purchase issue.”Daniel C.
“Whoa! It’s like Spotify but for academic articles.”@Phil_Robichaud
“I must say, @deepdyve is a fabulous solution to the independent researcher's problem of #access to #information.”@deepthiw
“My last article couldn't be possible without the platform @deepdyve that makes journal papers cheaper.”@JoseServera