Wavelet noise is an alternative to Perlin noise which reduces the problems of aliasing and detail loss that are encountered when Perlin noise is summed into a fractal. == Algorithm detail == The basic algorithm for 2-dimensional wavelet noise is as follows: Create an image, R {\displaystyle R} , filled with uniform white noise. Downsample R {\displaystyle R} to half-size to create R ↓ {\displaystyle R^{\downarrow }} , then upsample it back up to full size to create R ↓↑ {\displaystyle R^{\downarrow \uparrow }} . Subtract R ↓↑ {\displaystyle R^{\downarrow \uparrow }} from R {\displaystyle R} to create the end result, N {\displaystyle N} . This results in an image that contains all the information that cannot be represented at half-scale. From here, N {\displaystyle N} can be used similarly to Perlin noise to create fractal patterns.
Sentence extraction
Sentence extraction is a technique used for automatic summarization of a text. In this shallow approach, statistical heuristics are used to identify the most salient sentences of a text. Sentence extraction is a low-cost approach compared to more knowledge-intensive deeper approaches which require additional knowledge bases such as ontologies or linguistic knowledge. In short, sentence extraction works as a filter that allows only meaningful sentences to pass. The major downside of applying sentence-extraction techniques to the task of summarization is the loss of coherence in the resulting summary. Nevertheless, sentence extraction summaries can give valuable clues to the main points of a document and are frequently sufficiently intelligible to human readers. == Procedure == Usually, a combination of heuristics is used to determine the most important sentences within the document. Each heuristic assigns a (positive or negative) score to the sentence. After all heuristics have been applied, the highest-scoring sentences are included in the summary. The individual heuristics are weighted according to their importance. === Early approaches and some sample heuristics === Seminal papers which laid the foundations for many techniques used today have been published by Hans Peter Luhn in 1958 and H. P Edmundson in 1969. Luhn proposed to assign more weight to sentences at the beginning of the document or a paragraph. Edmundson stressed the importance of title-words for summarization and was the first to employ stop-lists in order to filter uninformative words of low semantic content (e.g. most grammatical words such as of, the, a). He also distinguished between bonus words and stigma words, i.e. words that probably occur together with important (e.g. the word form significant) or unimportant information. His idea of using key-words, i.e. words which occur significantly frequently in the document, is still one of the core heuristics of today's summarizers. With large linguistic corpora available today, the tf–idf value which originated in information retrieval, can be successfully applied to identify the key words of a text: If for example the word cat occurs significantly more often in the text to be summarized (TF = "term frequency") than in the corpus (IDF means "inverse document frequency"; here the corpus is meant by document), then cat is likely to be an important word of the text; the text may in fact be a text about cats.
Seq2seq
Seq2seq is a family of machine learning approaches used for natural language processing. Originally developed by Lê Viết Quốc, a Vietnamese computer scientist and a machine learning pioneer at Google Brain, this framework has become foundational in many modern AI systems. Applications include language translation, image captioning, conversational models, speech recognition, and text summarization. Seq2seq uses sequence transformation: it turns one sequence into another sequence. == History == One naturally wonders if the problem of translation could conceivably be treated as a problem in cryptography. When I look at an article in Russian, I say: 'This is really written in English, but it has been coded in some strange symbols. I will now proceed to decode. seq2seq is an approach to machine translation (or more generally, sequence transduction) with roots in information theory, where communication is understood as an encode-transmit-decode process, and machine translation can be studied as a special case of communication. This viewpoint was elaborated, for example, in the noisy channel model of machine translation. In practice, seq2seq maps an input sequence into a real-numerical vector by using a neural network (the encoder), and then maps it back to an output sequence using another neural network (the decoder). The idea of encoder-decoder sequence transduction had been developed in the early 2010s. The papers most commonly cited as the originators that produced seq2seq are two papers from 2014. In the seq2seq as proposed by them, both the encoder and the decoder were LSTMs. This had the "bottleneck" problem, since the encoding vector has a fixed size, so for long input sequences, information would tend to be lost, as they are difficult to fit into the fixed-length encoding vector. The attention mechanism, proposed in 2014, resolved the bottleneck problem. They called their model RNNsearch, as it "emulates searching through a source sentence during decoding a translation". A problem with seq2seq models at this point was that recurrent neural networks are difficult to parallelize. The 2017 publication of Transformers resolved the problem by replacing the encoding RNN with self-attention Transformer blocks ("encoder blocks"), and the decoding RNN with cross-attention causally-masked Transformer blocks ("decoder blocks"). === Priority dispute === One of the papers cited as the originator for seq2seq is (Sutskever et al 2014), published at Google Brain while they were on Google's machine translation project. The research allowed Google to overhaul Google Translate into Google Neural Machine Translation in 2016. Tomáš Mikolov claims to have developed the idea (before joining Google Brain) of using a "neural language model on pairs of sentences... and then [generating] translation after seeing the first sentence"—which he equates with seq2seq machine translation, and to have mentioned the idea to Ilya Sutskever and Quoc Le (while at Google Brain), who failed to acknowledge him in their paper. Mikolov had worked on RNNLM (using RNN for language modelling) for his PhD thesis, and is more notable for developing word2vec. == Architecture == The main reference for this section is. === Encoder === The encoder is responsible for processing the input sequence and capturing its essential information, which is stored as the hidden state of the network and, in a model with attention mechanism, a context vector. The context vector is the weighted sum of the input hidden states and is generated for every time instance in the output sequences. === Decoder === The decoder takes the context vector and hidden states from the encoder and generates the final output sequence. The decoder operates in an autoregressive manner, producing one element of the output sequence at a time. At each step, it considers the previously generated elements, the context vector, and the input sequence information to make predictions for the next element in the output sequence. Specifically, in a model with attention mechanism, the context vector and the hidden state are concatenated together to form an attention hidden vector, which is used as an input for the decoder. The seq2seq method developed in the early 2010s uses two neural networks: an encoder network converts an input sentence into numerical vectors, and a decoder network converts those vectors to sentences in the target language. The Attention mechanism was grafted onto this structure in 2014 and is shown below. Later it was refined into the encoder-decoder Transformer architecture of 2017. === Training vs prediction === There is a subtle difference between training and prediction. During training time, both the input and the output sequences are known. During prediction time, only the input sequence is known, and the output sequence must be decoded by the network itself. Specifically, consider an input sequence x 1 : n {\displaystyle x_{1:n}} and output sequence y 1 : m {\displaystyle y_{1:m}} . The encoder would process the input x 1 : n {\displaystyle x_{1:n}} step by step. After that, the decoder would take the output from the encoder, as well as the
Statistical shape analysis
Statistical shape analysis is an analysis of the geometrical properties of some given set of shapes by statistical methods. For instance, it could be used to quantify differences between male and female gorilla skull shapes, normal and pathological bone shapes, leaf outlines with and without herbivory by insects, etc. Important aspects of shape analysis are to obtain a measure of distance between shapes, to estimate mean shapes from (possibly random) samples, to estimate shape variability within samples, to perform clustering and to test for differences between shapes. One of the main methods used is principal component analysis (PCA). Statistical shape analysis has applications in various fields, including medical imaging, computer vision, computational anatomy, sensor measurement, and geographical profiling. == Landmark-based techniques == In the point distribution model, a shape is determined by a finite set of coordinate points, known as landmark points. These landmark points often correspond to important identifiable features such as the corners of the eyes. Once the points are collected some form of registration is undertaken. This can be a baseline methods used by Fred Bookstein for geometric morphometrics in anthropology. Or an approach like Procrustes analysis which finds an average shape. David George Kendall investigated the statistical distribution of the shape of triangles, and represented each triangle by a point on a sphere. He used this distribution on the sphere to investigate ley lines and whether three stones were more likely to be co-linear than might be expected. Statistical distribution like the Kent distribution can be used to analyse the distribution of such spaces. Alternatively, shapes can be represented by curves or surfaces representing their contours, by the spatial region they occupy. == Shape deformations == Differences between shapes can be quantified by investigating deformations transforming one shape into another. In particular a diffeomorphism preserves smoothness in the deformation. This was pioneered in D'Arcy Thompson's On Growth and Form before the advent of computers. Deformations can be interpreted as resulting from a force applied to the shape. Mathematically, a deformation is defined as a mapping from a shape x to a shape y by a transformation function Φ {\displaystyle \Phi } , i.e., y = Φ ( x ) {\displaystyle y=\Phi (x)} . Given a notion of size of deformations, the distance between two shapes can be defined as the size of the smallest deformation between these shapes. Diffeomorphometry is the focus on comparison of shapes and forms with a metric structure based on diffeomorphisms, and is central to the field of Computational anatomy. Diffeomorphic registration, introduced in the 90's, is now an important player with existing codes bases organized around ANTS, DARTEL, DEMONS, LDDMM, StationaryLDDMM, and FastLDDMM are examples of actively used computational codes for constructing correspondences between coordinate systems based on sparse features and dense images. Voxel-based morphometry (VBM) is an important technology built on many of these principles. Methods based on diffeomorphic flows are also used. For example, deformations could be diffeomorphisms of the ambient space, resulting in the LDDMM (Large Deformation Diffeomorphic Metric Mapping) framework for shape comparison.
Region Based Convolutional Neural Networks
Region-based Convolutional Neural Networks (R-CNN) are a family of machine learning models for computer vision, and specifically object detection and localization. The original goal of R-CNN was to take an input image and produce a set of bounding boxes as output, where each bounding box contains an object and also the category (e.g. car or pedestrian) of the object. In general, R-CNN architectures perform selective search over feature maps outputted by a CNN. R-CNN has been extended to perform other computer vision tasks, such as: tracking objects from a drone-mounted camera, locating text in an image, and enabling object detection in Google Lens. Mask R-CNN is also one of seven tasks in the MLPerf Training Benchmark, which is a competition to speed up the training of neural networks. == History == The following covers some of the versions of R-CNN that have been developed. November 2013: R-CNN. April 2015: Fast R-CNN. June 2015: Faster R-CNN. March 2017: Mask R-CNN. December 2017: Cascade R-CNN is trained with increasing Intersection over Union (IoU, also known as the Jaccard index) thresholds, making each stage more selective against nearby false positives. June 2019: Mesh R-CNN adds the ability to generate a 3D mesh from a 2D image. == Architecture == For review articles see. === Selective search === Given an image (or an image-like feature map), selective search (also called Hierarchical Grouping) first segments the image by the algorithm in (Felzenszwalb and Huttenlocher, 2004), then performs the following: Input: (colour) image Output: Set of object location hypotheses L Segment image into initial regions R = {r1, ..., rn} using Felzenszwalb and Huttenlocher (2004) Initialise similarity set S = ∅ foreach Neighbouring region pair (ri, rj) do Calculate similarity s(ri, rj) S = S ∪ s(ri, rj) while S ≠ ∅ do Get highest similarity s(ri, rj) = max(S) Merge corresponding regions rt = ri ∪ rj Remove similarities regarding ri: S = S \ s(ri, r∗) Remove similarities regarding rj: S = S \ s(r∗, rj) Calculate similarity set St between rt and its neighbours S = S ∪ St R = R ∪ rt Extract object location boxes L from all regions in R === R-CNN === With R-CNN, prediction follows a two-step process. A preprocessing selective search step generates a large set of candidate objects (typically as many as 2000), known as regions of interest (ROI). These are forwarded to a CNN, which predicts an object class score and bounding box estimate, independently for each ROI. Importantly, the ROIs are heavily filtered to remove excess candidates. This is achieved using two mechanism. Filtering begins by removing ROIs assigned to the background category. This is a specialized category, which is scored by the CNN alongside other categories. An unfortunate reality is that remaining ROIs typically suffer from heavy duplication. Namely, multiple ROIs that cover same objects in the image are all assigned non-background categories. This is resolved by a heuristic non-maximum suppression (NMS) step. === Fast R-CNN === While the original R-CNN independently computed the neural network features on each of as many as two thousand regions of interest, Fast R-CNN runs the neural network once on the whole image. At the end of the network is a ROIPooling module, which slices out each ROI from the network's output tensor, reshapes it, and classifies it. As in the original R-CNN, the Fast R-CNN uses selective search to generate its region proposals. === Faster R-CNN === While Fast R-CNN used selective search to generate ROIs, Faster R-CNN integrates the ROI generation into the neural network itself. === Mask R-CNN === While previous versions of R-CNN focused on object detections, Mask R-CNN adds instance segmentation. Mask R-CNN also replaced ROIPooling with a new method called ROIAlign, which can represent fractions of a pixel.
Microscope image processing
Microscope image processing is a broad term that covers the use of digital image processing techniques to process, analyze and present images obtained from a microscope. Such processing is now commonplace in a number of diverse fields such as medicine, biological research, cancer research, drug testing, metallurgy, etc. A number of manufacturers of microscopes now specifically design in features that allow the microscopes to interface to an image processing system. == Image acquisition == Until the early 1990s, most image acquisition in video microscopy applications was typically done with an analog video camera, often simply closed circuit TV cameras. While this required the use of a frame grabber to digitize the images, video cameras provided images at full video frame rate (25-30 frames per second) allowing live video recording and processing. While the advent of solid state detectors yielded several advantages, the real-time video camera was actually superior in many respects. Today, acquisition is usually done using a CCD camera mounted in the optical path of the microscope. The camera may be full colour or monochrome. Very often, very high resolution cameras are employed to gain as much direct information as possible. Cryogenic cooling is also common, to minimise noise. Often digital cameras used for this application provide pixel intensity data to a resolution of 12-16 bits, much higher than is used in consumer imaging products. Ironically, in recent years, much effort has been put into acquiring data at video rates, or higher (25-30 frames per second or higher). What was once easy with off-the-shelf video cameras now requires special, high speed electronics to handle the vast digital data bandwidth. Higher speed acquisition allows dynamic processes to be observed in real time, or stored for later playback and analysis. Combined with the high image resolution, this approach can generate vast quantities of raw data, which can be a challenge to deal with, even with a modern computer system. While current CCD detectors allow very high image resolution, often this involves a trade-off because, for a given chip size, as the pixel count increases, the pixel size decreases. As the pixels get smaller, their well depth decreases, reducing the number of electrons that can be stored. In turn, this results in a poorer signal-to-noise ratio. For best results, one must select an appropriate sensor for a given application. Because microscope images have an intrinsic limiting resolution, it often makes little sense to use a noisy, high resolution detector for image acquisition. A more modest detector, with larger pixels, can often produce much higher quality images because of reduced noise. This is especially important in low-light applications such as fluorescence microscopy. Moreover, one must also consider the temporal resolution requirements of the application. A lower resolution detector will often have a significantly higher acquisition rate, permitting the observation of faster events. Conversely, if the observed object is motionless, one may wish to acquire images at the highest possible spatial resolution without regard to the time required to acquire a single image. == 2D image techniques == Image processing for microscopy application begins with fundamental techniques intended to most accurately reproduce the information contained in the microscopic sample. This might include adjusting the brightness and contrast of the image, averaging images to reduce image noise and correcting for illumination non-uniformities. Such processing involves only basic arithmetic operations between images (i.e. addition, subtraction, multiplication and division). The vast majority of processing done on microscope image is of this nature. Another class of common 2D operations called image convolution are often used to reduce or enhance image details. Such "blurring" and "sharpening" algorithms in most programs work by altering a pixel's value based on a weighted sum of that and the surrounding pixels (a more detailed description of kernel based convolution deserves an entry for itself) or by altering the frequency domain function of the image using Fourier Transform. Most image processing techniques are performed in the Frequency domain. Other basic two dimensional techniques include operations such as image rotation, warping, color balancing etc. At times, advanced techniques are employed with the goal of "undoing" the distortion of the optical path of the microscope, thus eliminating distortions and blurring caused by the instrumentation. This process is called deconvolution, and a variety of algorithms have been developed, some of great mathematical complexity. The end result is an image far sharper and clearer than could be obtained in the optical domain alone. This is typically a 3-dimensional operation, that analyzes a volumetric image (i.e. images taken at a variety of focal planes through the sample) and uses this data to reconstruct a more accurate 3-dimensional image. == 3D image techniques == Another common requirement is to take a series of images at a fixed position, but at different focal depths. Since most microscopic samples are essentially transparent, and the depth of field of the focused sample is exceptionally narrow, it is possible to capture images "through" a three-dimensional object using 2D equipment like confocal microscopes. Software is then able to reconstruct a 3D model of the original sample which may be manipulated appropriately. The processing turns a 2D instrument into a 3D instrument, which would not otherwise exist. In recent times this technique has led to a number of scientific discoveries in cell biology. == Analysis == Analysis of images will vary considerably according to application. Typical analysis includes determining where the edges of an object are, counting similar objects, calculating the area, perimeter length and other useful measurements of each object. A common approach is to create an image mask which only includes pixels that match certain criteria, then perform simpler scanning operations on the resulting mask. It is also possible to label objects and track their motion over a series of frames in a video sequence.
Weight initialization
In deep learning, weight initialization or parameter initialization describes the initial step in creating a neural network. A neural network contains trainable parameters that are modified during training: weight initialization is the pre-training step of assigning initial values to these parameters. The choice of weight initialization method affects the speed of convergence, the scale of neural activation within the network, the scale of gradient signals during backpropagation, and the quality of the final model. Proper initialization is necessary for avoiding issues such as vanishing and exploding gradients and activation function saturation. Note that even though this article is titled "weight initialization", both weights and biases are used in a neural network as trainable parameters, so this article describes how both of these are initialized. Similarly, trainable parameters in convolutional neural networks (CNNs) are called kernels and biases, and this article also describes these. == Constant initialization == We discuss the main methods of initialization in the context of a multilayer perceptron (MLP). Specific strategies for initializing other network architectures are discussed in later sections. For an MLP, there are only two kinds of trainable parameters, called weights and biases. Each layer l {\displaystyle l} contains a weight matrix W ( l ) ∈ R n l − 1 × n l {\displaystyle W^{(l)}\in \mathbb {R} ^{n_{l-1}\times n_{l}}} and a bias vector b ( l ) ∈ R n l {\displaystyle b^{(l)}\in \mathbb {R} ^{n_{l}}} , where n l {\displaystyle n_{l}} is the number of neurons in that layer. A weight initialization method is an algorithm for setting the initial values for W ( l ) , b ( l ) {\displaystyle W^{(l)},b^{(l)}} for each layer l {\displaystyle l} . The simplest form is zero initialization: W ( l ) = 0 , b ( l ) = 0 {\displaystyle W^{(l)}=0,b^{(l)}=0} Zero initialization is usually used for initializing biases, but it is not used for initializing weights, as it leads to symmetry in the network, causing all neurons to learn the same features. In this page, we assume b = 0 {\displaystyle b=0} unless otherwise stated. Recurrent neural networks typically use activation functions with bounded range, such as sigmoid and tanh, since unbounded activation may cause exploding values. (Le, Jaitly, Hinton, 2015) suggested initializing weights in the recurrent parts of the network to identity and zero bias, similar to the idea of residual connections and LSTM with no forget gate. In most cases, the biases are initialized to zero, though some situations can use a nonzero initialization. For example, in multiplicative units, such as the forget gate of LSTM, the bias can be initialized to 1 to allow good gradient signal through the gate. For neurons with ReLU activation, one can initialize the bias to a small positive value like 0.1, so that the gradient is likely nonzero at initialization, avoiding the dying ReLU problem. == Random initialization == Random initialization means sampling the weights from a normal distribution or a uniform distribution, usually independently. === LeCun initialization === LeCun initialization, popularized in (LeCun et al., 1998), is designed to preserve the variance of neural activations during the forward pass. It samples each entry in W ( l ) {\displaystyle W^{(l)}} independently from a distribution with mean 0 and variance 1 / n l − 1 {\displaystyle 1/n_{l-1}} . For example, if the distribution is a continuous uniform distribution, then the distribution is U ( ± 3 / n l − 1 ) {\displaystyle {\mathcal {U}}(\pm {\sqrt {3/n_{l-1}}})} . === Glorot initialization === Glorot initialization (or Xavier initialization) was proposed by Xavier Glorot and Yoshua Bengio. It was designed as a compromise between two goals: to preserve activation variance during the forward pass and to preserve gradient variance during the backward pass. For uniform initialization, it samples each entry in W ( l ) {\displaystyle W^{(l)}} independently and identically from U ( ± 6 / ( n l + 1 + n l − 1 ) ) {\displaystyle {\mathcal {U}}(\pm {\sqrt {6/(n_{l+1}+n_{l-1})}})} . In the context, n l − 1 {\displaystyle n_{l-1}} is also called the "fan-in", and n l + 1 {\displaystyle n_{l+1}} the "fan-out". When the fan-in and fan-out are equal, then Glorot initialization is the same as LeCun initialization. === He initialization === As Glorot initialization performs poorly for ReLU activation, He initialization (or Kaiming initialization) was proposed by Kaiming He et al. for networks with ReLU activation. It samples each entry in W ( l ) {\displaystyle W^{(l)}} from N ( 0 , 2 / n l − 1 ) {\displaystyle {\mathcal {N}}(0,2/n_{l-1})} . === Orthogonal initialization === (Saxe et al. 2013) proposed orthogonal initialization: initializing weight matrices as uniformly random (according to the Haar measure) semi-orthogonal matrices, multiplied by a factor that depends on the activation function of the layer. It was designed so that if one initializes a deep linear network this way, then its training time until convergence is independent of depth. Sampling a uniformly random semi-orthogonal matrix can be done by initializing X {\displaystyle X} by IID sampling its entries from a standard normal distribution, then calculate ( X X ⊤ ) − 1 / 2 X {\displaystyle \left(XX^{\top }\right)^{-1/2}X} or its transpose, depending on whether X {\displaystyle X} is tall or wide. For CNN kernels with odd widths and heights, orthogonal initialization is done this way: initialize the central point by a semi-orthogonal matrix, and fill the other entries with zero. As an illustration, a kernel K {\displaystyle K} of shape 3 × 3 × c × c ′ {\displaystyle 3\times 3\times c\times c'} is initialized by filling K [ 2 , 2 , : , : ] {\displaystyle K[2,2,:,:]} with the entries of a random semi-orthogonal matrix of shape c × c ′ {\displaystyle c\times c'} , and the other entries with zero. (Balduzzi et al., 2017) used it with stride 1 and zero-padding. This is sometimes called the Orthogonal Delta initialization. Related to this approach, unitary initialization proposes to parameterize the weight matrices to be unitary matrices, with the result that at initialization they are random unitary matrices (and throughout training, they remain unitary). This is found to improve long-sequence modelling in LSTM. Orthogonal initialization has been generalized to layer-sequential unit-variance (LSUV) initialization. It is a data-dependent initialization method, and can be used in convolutional neural networks. It first initializes weights of each convolution or fully connected layer with orthonormal matrices. Then, proceeding from the first to the last layer, it runs a forward pass on a random minibatch, and divides the layer's weights by the standard deviation of its output, so that its output has variance approximately 1. === Fixup initialization === In 2015, the introduction of residual connections allowed very deep neural networks to be trained, much deeper than the ~20 layers of the previous state of the art (such as the VGG-19). Residual connections gave rise to their own weight initialization problems and strategies. These are sometimes called "normalization-free" methods, since using residual connection could stabilize the training of a deep neural network so much that normalizations become unnecessary. Fixup initialization is designed specifically for networks with residual connections and without batch normalization, as follows: Initialize the classification layer and the last layer of each residual branch to 0. Initialize every other layer using a standard method (such as He initialization), and scale only the weight layers inside residual branches by L − 1 2 m − 2 {\displaystyle L^{-{\frac {1}{2m-2}}}} . Add a scalar multiplier (initialized at 1) in every branch and a scalar bias (initialized at 0) before each convolution, linear, and element-wise activation layer. Similarly, T-Fixup initialization is designed for Transformers without layer normalization. === Others === Instead of initializing all weights with random values on the order of O ( 1 / n ) {\displaystyle O(1/{\sqrt {n}})} , sparse initialization initialized only a small subset of the weights with larger random values, and the other weights zero, so that the total variance is still on the order of O ( 1 ) {\displaystyle O(1)} . Random walk initialization was designed for MLP so that during backpropagation, the L2 norm of gradient at each layer performs an unbiased random walk as one moves from the last layer to the first. Looks linear initialization was designed to allow the neural network to behave like a deep linear network at initialization, since W R e L U ( x ) − W R e L U ( − x ) = W x {\displaystyle W\;\mathrm {ReLU} (x)-W\;\mathrm {ReLU} (-x)=Wx} . It initializes a matrix W {\displaystyle W} of shape R n 2 × m {\displaystyle \mathbb {R} ^{{\frac {n}{2}}\times m}} by any method, such as orthogonal initialization, t