Quantitative Biology > Neurons and Cognition
[Submitted on 19 Aug 2016 (v1), last revised 13 Jan 2017 (this version, v2)]
Title:Disambiguating the role of noise correlations when decoding neural populations together
View PDFAbstract:One of the most controversial problems in neural decoding is quantifying the information loss caused by ignoring noise correlations during optimal brain computations. For more than a decade, the measure here called $ \Delta I^{DL} $ has been believed exact. However, we have recently shown that it can exceed the information loss $ \Delta I^{B} $ caused by optimal decoders constructed ignoring noise correlations. Unfortunately, the different information notions underlying $ \Delta I^{DL} $ and $ \Delta I^{B} $, and the putative rigorous information-theoretical derivation of $ \Delta I^{DL} $, both render unclear whether those findings indicate either flaws in $ \Delta I^{DL} $ or major departures from traditional relations between information and decoding. Here we resolve this paradox and prove that, under certain conditions, observing $ \Delta I^{DL} {>}\Delta I^{B} $ implies that $ \Delta I^{DL} $ is flawed. Motivated by this analysis, we test both measures using neural populations that transmit independent information. Our results show that $ \Delta I^{DL} $ may deem noise correlations more important when decoding the populations together than when decoding them in parallel, whereas the opposite may occur for $ \Delta I^{B} $. We trace these phenomena back, for $ \Delta I^{B} $, to the choice of tie-breaking rules, and for $ \Delta I^{DL} $, to unforeseen limitations within its information-theoretical foundations. Our study contributes with better estimates that potentially improve theoretical and experimental inferences currently drawn from $ \Delta I^{DL} $ without noticing that it may constitute an upper bound. On the practical side, our results promote the design of optimal decoding algorithms and neuroprosthetics without recording noise correlations, thereby saving experimental and computational resources.
Submission history
From: Hugo Gabriel Eyherabide Dr [view email][v1] Fri, 19 Aug 2016 06:00:22 UTC (2,989 KB)
[v2] Fri, 13 Jan 2017 10:22:28 UTC (4,430 KB)
Current browse context:
q-bio
References & Citations
Bibliographic and Citation Tools
Bibliographic Explorer (What is the Explorer?)
Litmaps (What is Litmaps?)
scite Smart Citations (What are Smart Citations?)
Code, Data and Media Associated with this Article
CatalyzeX Code Finder for Papers (What is CatalyzeX?)
DagsHub (What is DagsHub?)
Gotit.pub (What is GotitPub?)
Papers with Code (What is Papers with Code?)
ScienceCast (What is ScienceCast?)
Demos
Recommenders and Search Tools
Influence Flower (What are Influence Flowers?)
Connected Papers (What is Connected Papers?)
CORE Recommender (What is CORE?)
arXivLabs: experimental projects with community collaborators
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.