CNS*2020 Online has ended
Welcome to the Sched instance for CNS*2020 Online! Please read the instruction document on detailed information on CNS*2020.
Back To Schedule
Monday, July 20 • 7:00pm - 8:00pm
P1: Homeostatic recognition circuits emulating network-wide bursting and surprise

Log in to save this to your schedule, view media, leave feedback and see who's attending!

Feedback form is now closed.
Meeting link:

Tsvi Achler

Understanding the circuits of recognition is essential to build a deeper understanding of virtually all of the brains behaviors and circuits.

The goal of this work is to capture simultaneous findings on both the neural and behavioral levels, namely Network Wide Bursting (NWB) dynamics with surprise (unexpected inputs), using a hypothesized recognition circuit based on the idea of homeostasis flow.

If real neural brains at a resting state are presented with an unexpected or new stimulus, the brain network shows a fast network-wide increase in activation (NWB of many neurons) followed by a slower inhibition, until the network settles again to a resting state. Bursting phenomena during recognition is found ubiquitously in virtually every type of organism, within isolated brain dissections and even neural tissue grown in a dish (Fig. 1). Its source and function remain poorly understood. Behavioral manifestation of surprise can be observed if the input is much unexpected and may involve multiple brain regions.

The homeostatic flow model posits that activation from inputs is balanced with top down pre-synaptic regulatory feedback from output neurons. Information is projected from inputs to outputs with forward connections then back to inputs with backwards homeostatic connections which inhibits the inputs. This effectively acts to balance the inputs & outputs (homeostasis) and generates an internal error-dependent input. This homeostatic input is then projected again to outputs and back again until output values relate recognition. This occurs during recognition and no weights are learned.

When a surprise or unexpected input stimulus is presented, NWB occurs because the homeostatic balance is disturbed with the new stimulus. The system subsequently calms down as it settles back to a new homeostasis.

In comparing to existing models, this circuit is different from Adaptive Resonance Theory because: 1) no lateral connections are required (inhibitory or otherwise) 2) all neurons feed backwards pre-synaptically at the same time 3) there is no vigilance parameter. It is different from Hopfield networks because instead of top-down feedback being positive, it is negative (inhibitory & homeostatic). This changes the functions and dynamics of the model making it stable: its dynamics eventually converge to steady state as long as inputs do not change.

The homeostatic feedback should not be confused with error of learning algorithms since: 1) it is implemented during recognition 2) does not adjust any weights at any time 3) not generated using training data. It is different from generative and predictive coding models because 1) it is primarily used during recognition not learning 2) the generative and recognition components are inseparable and contained within a single integrated homeostatic circuit.

The network is connectionist but approximates a Bayesian network by: 1) homeostatic weights are roughly equivalent to Bayesian likelihood values 2) output values can behave as Bayesian priors if they are maintained externally or if inputs suddenly change. Maintaining priors changes circuit recognition and dynamics without changing weights.

Learning can achieved with simple Hebbian learning, obtaining weights that are similar to Bayesian likelyhood. Both directions of the homeostatic process learn the same weights. Single layer learning is demonstrated with standard MNIST digits while capturing the neural findings of NWB.


avatar for Tsvi Achler

Tsvi Achler

Chief Science Officer, Optimizing Mind

Monday July 20, 2020 7:00pm - 8:00pm CEST
Slot 20