$$\text{lr}_t := \mathrm{lr} \cdot \frac{\sqrt{1 - \beta_2^t}}{1 - \beta_1^t}$$
$$m_t := \beta_1 \cdot m_{t-1} + (1 - \beta_1) \cdot g$$
$$v_t := \beta_2 \cdot v_{t-1} + (1 - \beta_2) \cdot g^2$$
$$\text{var} := \begin{cases} \text{var} - (m_t \beta_1 + g \cdot (1 - \beta_1))\cdot\text{lr}_t/(\sqrt{v_t} + \epsilon), &\text{if use_nesterov}\\\\  \text{var} - m_t \cdot \text{lr}_t /(\sqrt{v_t} + \epsilon), &\text{otherwise} \end{cases}$$
  }];

the breadths of the Fringes seem'd to be in the progression of the
Numbers 1, sqrt(1/3), sqrt(1/5), and their Intervals to be in the
same progression with them; that is, the Fringes and their Intervals
together to be in the continual progression of the Numbers 1,
sqrt(1/2), sqrt(1/3), sqrt(1/4), sqrt(1/5), or thereabouts. And
these Proportions held the same very nearly at all distances from the

		{"Signbit", Func, 0},
		{"Sin", Func, 0},
		{"Sincos", Func, 0},
		{"Sinh", Func, 0},
		{"SmallestNonzeroFloat32", Const, 0},
		{"SmallestNonzeroFloat64", Const, 0},
		{"Sqrt", Func, 0},
		{"Sqrt2", Const, 0},
		{"SqrtE", Const, 0},
		{"SqrtPhi", Const, 0},
		{"SqrtPi", Const, 0},
		{"Tan", Func, 0},
		{"Tanh", Func, 0},
		{"Trunc", Func, 0},
		{"Y0", Func, 0},
		{"Y1", Func, 0},

		return true
	}
	// match: (Cvt64Fto32F sqrt0:(Sqrt (Cvt32Fto64F x)))
	// cond: sqrt0.Uses==1
	// result: (Sqrt32 x)
	for {
		sqrt0 := v_0
		if sqrt0.Op != OpSqrt {
			break
		}
		sqrt0_0 := sqrt0.Args[0]
		if sqrt0_0.Op != OpCvt32Fto64F {
			break
		}
		x := sqrt0_0.Args[0]
		if !(sqrt0.Uses == 1) {
			break
		}
		v.reset(OpSqrt32)
		v.AddArg(x)

        1e-10))` Alternatively, you can override `convolution_op`: `python class
        StandardizedConv2D(tf.keras.Layer): def convolution_op(self, inputs,
        kernel): mean, var = tf.nn.moments(kernel, axes=[0, 1, 2],
        keepdims=True) # Author code uses std + 1e-5 return
        super().convolution_op(inputs, (kernel - mean) / tf.sqrt(var + 1e-10))`