[Paper Review] Topological Effects of Synaptic Time Dependent Plasticity

James Kozloski, Guillermo Cecchi|arXiv (Cornell University)|Sep 30, 2008

Neural dynamics and brain function21 references21 citations

TL;DR

This paper proposes that Spike Timing-Dependent Plasticity (STDP) acts as a universal loop-regulating mechanism in neural networks, minimizing functional loops through anti-symmetric synaptic weight changes based on pre- and post-synaptic spike timing. The key result is a rigorous analytical and simulation-based proof that STDP drives networks toward feed-forward topology by weakening recurrent connections, with implications for microcircuit and global brain network organization.

ABSTRACT

We show that the local Spike Timing-Dependent Plasticity (STDP) rule has the effect of regulating the trans-synaptic weights of loops of any length within a simulated network of neurons. We show that depending on STDP's polarity, functional loops are formed or eliminated in networks driven to normal spiking conditions by random, partially correlated inputs, where functional loops comprise weights that exceed a non-zero threshold. We further prove that STDP is a form of loop-regulating plasticity for the case of a linear network comprising random weights drawn from certain distributions. Thus a notable local synaptic learning rule makes a specific prediction about synapses in the brain in which standard STDP is present: that under normal spiking conditions, they should participate in predominantly feed-forward connections at all scales. Our model implies that any deviations from this prediction would require a substantial modification to the hypothesized role for standard STDP. Given its widespread occurrence in the brain, we predict that STDP could also regulate long range synaptic loops among individual neurons across all brain scales, up to, and including, the scale of global brain network topology.

Motivation & Objective

• To investigate whether STDP, a local synaptic plasticity rule, can regulate the topology of neural microcircuits, particularly the formation and elimination of functional loops.
• To determine if STDP's anti-symmetric learning rule inherently suppresses recurrent connectivity, even in polysynaptic loops, under normal spiking conditions.
• To provide a formal mathematical proof that STDP acts as a loop-regulating plasticity mechanism in linear networks with random, Gaussian-distributed inputs.
• To test the hypothesis that STDP enforces a predominantly feed-forward architecture across all brain scales, from microcircuits to global network topology.
• To establish that deviations from feed-forward topology in real neural circuits would necessitate a revision of the standard STDP hypothesis.

Proposed method

• Formal derivation of the STDP learning rule in the adiabatic approximation using a correlator-based update: Δw_xy ∝ ∫ C_xy(t) S(t) dt, where S(t) is anti-symmetric and C_xy(t) is the cross-correlation of pre- and post-synaptic activity.
• Modeling the network as a linear dynamical system ẋ(t) = Wx(t) + ξ(t), with uncorrelated Gaussian noise inputs satisfying ⟨ξ_i(t)ξ_j^T(t+τ)⟩ = σ²δ(τ)δ_ij.
• Deriving the steady-state correlator C_0 via the Lyapunov equation: W^T C_0 + C_0 W = -Q Q^T, with Q Q^T = I for homogeneous noise.
• Analyzing the update of the weight matrix ΔW using the formal solution C_0 = ∫₀^∞ e^{W^T t} e^{W t} dt for stable W (all eigenvalues with negative real parts).
• Proving that the change in a Lyapunov-like energy function ΔU is non-positive by showing tr[K₁] ≥ 0 and tr[K₂] ≥ 0, implying loop suppression and increased dynamical stability.
• Using eigenvalue decomposition and matrix trace identities to show δU ≤ 0 under first-order approximation, confirming that STDP reduces loopiness and enhances system stability.

Experimental results

Research questions

• RQ1 Does STDP inherently suppress the formation of functional loops in neural networks, even in polysynaptic feedback configurations?
• RQ2 Can STDP be mathematically proven to act as a loop-regulating plasticity rule in linear networks with random, uncorrelated inputs?
• RQ3 What is the relationship between STDP-induced synaptic changes and the stability of neural network dynamics?
• RQ4 To what extent does STDP enforce a feed-forward architecture in neural microcircuits, and what are the implications for global brain network topology?
• RQ5 Would observed recurrent connectivity in real neural circuits contradict the standard hypothesis of STDP as a primary organizer of neural connectivity?

Key findings

• STDP acts as a loop-regulating plasticity rule that systematically weakens feedback connections in both direct and polysynaptic loops, leading to the elimination of functional loops.
• The analytical proof shows that the change in a Lyapunov-like energy function ΔU is non-positive (ΔU ≤ 0), implying that STDP drives the network toward a state of minimal loopiness and maximal dynamical stability.
• For stable weight matrices W (all eigenvalues with negative real parts), the trace of the update term tr[K₂] is non-negative, confirming that STDP reduces loop formation and enhances stability.
• The first-order approximation of the weight update yields δU ∝ (tr[A²] - tr[AA^T]) ≤ 0, demonstrating that STDP reduces the system's loopiness and increases stability.
• The result is robust under the assumption of homogeneous noise (QQ^T = I), and the formal solution C_0 = ∫₀^∞ e^{W^T t} e^{W t} dt ensures the validity of the trace-based stability analysis.
• The study implies that any persistent recurrent connectivity in real neural circuits under normal spiking conditions would contradict the standard STDP hypothesis, requiring a revision of its role in circuit organization.