Designing Dynamical Properties of Brain–Machine Interfaces to Optimize Task-Specific Performance

Pretty heavy math, will have to go over this again. Premise is that small changes in closed-loop BMI dynamical properties can have significant and predicatable effects on performance. PVKF was used as a specific algorithm to study.

Two undesired effects:

  • presence of unintended attractor points which pulled the cursor to arbitrary workspace positions.
  • calibration-dependent control memory property which parameterized the tradeoff between speed and hold performance. likely driven by fluctuation in the closed-loop control memory of the decoder.

Both properties were not only present in experimental data but also their mathematical causes could be traced back to the mechanics of the closed-loop decoder. For KF, and generally for recursive Bayesian decoders, the closed-loop properties induced by interaction between the sate-space model and the neural firing model can produce unintentional dynamical artifacts.

There are numerous benefits of keeping both the decoder and the neural ensemble fixed across sessions. However, CLDA is needed to adapt to nonstationarity of the ensemble, and initial calibration. But CLDA can also change system properties that one might prefer to keep fixed if one wishes to keep the system as consistent as possible before and after calibration.

References to Read

  • F. R. Willett , A. J. Suminski , A. H. Fagg and N. G. Hatsopoulos “Improving brain-machine interfaceperformance by decoding intended future movements”, J.Neural Eng., vol. 10, no. 2, pp.026011 2013
  • B. Jarosiewicz , S. M. Chase , G. W. Fraser , M. Velliste , R. E. Kass and A. B. Schwartz “Functional network reorganizationduring learning in a brain-computer interface paradigm”, Proc. Nat. Acad. Sci., vol. 105, no. 49, pp.19486 -19491 2008
  • E. J. Hwang , P. M. Bailey and R. A. Andersen “Volitional control of neuralactivity relies on the natural motor repertoire”, Curr.Biol., vol. 23, no. 5, pp.353 -361 2013
  • W. Wu , Y. Gao , E. Bienenstock , J. P. Donoghue and M. J. Black “Bayesian population decoding of motor corticalactivity using a Kalman filter”, Neural Comput., vol. 18, no. 1, pp.80 -118 2006
  • Z. Li , J. E. O'Doherty , M. A. Lebedev and M. A. L. Nicolelis “Adaptive decoding for brain-machineinterfaces through Bayesian parameter updates”, NeuralComput., vol. 23, no. 12, pp.3162 -3204 2011
  • M. Golub , S. M. Chase and B. M. Yu “Learning an internal dynamics model fromcontrol demonstration”, Proc. Int. Conf. Mach.Learn., vol. 28, pp.606 -614 2013