Browsing by Subject "Decoding"

The retina signals visual information to the brain with the parallel channels of different retinal ganglion cell (RGC) types, whose signals ultimately lead to visual perception. Between cloudy nights and sunny days, the retina must combat the trillion-fold change in mean light intensity to successfully convey visual information. Critically, the nature of both signal and noise in RGC populations is altered across this broad range of light levels, creating a rich problem of how visual messages are encoded by the retina and transmitted to the brain. This thesis addresses these topics using large-scale multielectrode array recordings of RGC populations in different light conditions. In Chapter 2, I characterize how retinal signaling is altered over a wide range of light intensities. Chapter 3 investigates how adaptation impacts visual encoding of different RGC types. My results suggest that although retinal computations change substantially over light conditions, there are some elements of visual encoding that are invariant to light adaptation. Finally, Chapter 4 examines adaptation-induced changes in the structure of correlated activity and the subsequent impact on processing retinal output. The findings of this chapter clarify the nature of RGC responses crucial for downstream readout across light levels. Overall, this work identifies aspects of RGC activity that are important for encoding visual information and decoding retinal output from starlight to sunlight.

In this report we validate that Minimum Weight Perfect Matching (MWPM) decoding ofa surface code helps attain a higher code threshold than standard decoding by taking into account the correlations of errors that occur on the lattice. Correlated decoding cannot be directly performed using MWPM due to the presence of hyperedges in the graph. Decomposing correlated errors into simultaneous X and Z excitations, along with considering an asymmetric code for optimal performance of this scheme, yields a codecapacity threshold of 15% for standard depolarizing error by updating the weights of one decoding graph based on the corrections of another.