Markov Chain
2019-09-08
A Markov chain is a stochastic model describing a sequence of possible events in which the probability of each event depends only on the state attained in the previous event. (“memorylessness”)Continue Reading
Harmonic Mean Of Precision and Recall – F1 Score
2019-08-09
In statistical analysis of binary classification, the Harmonic mean in ML is a measure of a test’s accuracy and is known as F1 score or subcontrary mean. The F1 score (Harmonic mean) is combined by using the recall and the precision , and is better then taking the arithmetic meanContinue Reading
Q-learning – Model-free reinforcement learning algorithm
2019-07-06
Learns a policy which tells an agent what action to take under what circumstances. Q-learning learns a policy that is optimal in the sense that it maximizes the expected value of the total reward over any and all successive steps, starting from the current state.Continue Reading
Gradient
2019-06-09
Gradient is a vector that is made up from a the derivatives of multi-variate function. Gradient of an Image = ∇f = [∂f/∂x,∂f/∂y] The gradient of a function is the vector direction/angle of the most rapid increase in Intensity(Getting Brighter), and the magnitude of that vector is how much itContinue Reading
Feature extraction – representation of unstructured data in a structured way
2019-05-31
Images – A table of pixels, usually of RGB(3D columns – a.k.a a tensor) – can be ravled/unrolled/flattened to a 1D longer array. Sound – long series of numbers(1D array) when digitally recorded(x = time(duration column), y = amplitude of the sound) – usually represented by frequences. Text – partsContinue Reading
Reduce and Expand in Down/Up Sampling without Aliasing
2019-05-24
Reduce Operator – DownSampling without aliasing: Usually a 5 or 7 tap separable 1D filter like a (1,4,6,4,1) or a (1,6,15,20,15,6,1) filter is used. We need half the number of pixels: An easy example , each odd pixel gets 5 separately filtered by (1,4,6,4,1) filter. Reproducing / Expanding OperatorContinue Reading
Hierarchical Lucas and Kanada with multi-scale
2019-05-24
The Lucas–Kanade method is a widely used differential method for optical flow estimation. Assumptions – flow is essentially constant in a local neighbourhood – solves the basic optical flow equations for all the pixels in that neighbourhood, by the least squares criterion. motion is linear : When it is non linearContinue Reading
The aperture problem
2019-05-24
The motion direction of a contour is ambiguous(usually to a combination of left/right with up/down), because the motion component parallel to the line (edge) cannot be inferred based on the visual input. You can only tell the motion locally in the direction perpendicular to the edge(line).Continue Reading
Optical Flow Constraints
2019-05-18
In order to compute Optical Flow there are two assumptions/constraints: Image moves in u(x cord) and v(y cord): The brightness(grayscale)/color consistency is same between the original image I ( x, y,t), and the image in I ( x+u, y+v,t+1). The change in motion is almost zero(smooth), which means that the original IContinue Reading
Karatsuba algorithm – O(n^2) becomes O(n^1.58)
2019-04-25
The Karatsuba algorithm is a fast multiplication algorithm of two (works better for large) n-digit numbers(x and y),using three multiplications of smaller numbers(about half as many digits as x or y), plus some additions and digit shifts. The steps are: For any positive integer less than (string length): where and are less than .Continue Reading