softmax(QKᵀ/√dₖ)Viħ ∂ψ/∂t = ĤψH(p) = −Σ pᵢ log pᵢP(w₁…wₙ) = Π P(wᵢ | w₍₋ᵢ₎)∇·E = ρ/ε₀θ ← θ − η∇L(θ)e^{iπ} + 1 = 0σ(z) = 1/(1+e^{−z})S = k log Wλ = h/pargmax_θ log P(D|θ)E = mc²attention is all you needx ~ 𝒩(μ, σ²)D_KL(P ∥ Q) = Σ p log(p/q)P(θ|D) ∝ P(D|θ) P(θ)Δx Δp ≥ ħ/2f̂(ξ) = ∫ f(x) e^{−2πixξ} dxL = T − VPPL = 2^{H(p)}cos(u, v) = u·v / ‖u‖‖v‖PE₍pos,2i₎ = sin(pos/10000^{2i/d})∂²u/∂t² = c²∇²uf ∝ 1/rank (Zipf)ReLU(x) = max(0, x)∮ B·dl = μ₀(I + ε₀ dΦ/dt)