In computer programming, the exclusive or swap (sometimes shortened to XOR swap) is an algorithm that uses the exclusive or bitwise operation to swap the values of two variables without using the temporary variable which is normally required. The algorithm is primarily a novelty and a way of demonstrating properties of the exclusive or operation. It is sometimes discussed as a program optimization, but there are almost no cases where swapping via exclusive or provides benefit over the standard, obvious technique. == The algorithm == Conventional swapping requires the use of a temporary storage variable. Using the XOR swap algorithm, however, no temporary storage is needed. The algorithm is as follows: Since XOR is a commutative operation, either X XOR Y or Y XOR X can be used interchangeably in any of the foregoing three lines. Note that on some architectures the first operand of the XOR instruction specifies the target location at which the result of the operation is stored, preventing this interchangeability. The algorithm typically corresponds to three machine-code instructions, represented by corresponding pseudocode and assembly instructions in the three rows of the following table: In the above System/370 assembly code sample, R1 and R2 are distinct registers, and each XR operation leaves its result in the register named in the first argument. Using x86 assembly, values X and Y are in registers eax and ebx (respectively), and xor places the result of the operation in the first register (Note: x86 supports XCHG instruction so using triple XOR do not make sense on this architecture). In RISC-V assembly, value X and Y are in registers x10 and x11, and xor places the result of the operation in the first operand. However, in the pseudocode or high-level language version or implementation, the algorithm fails if x and y use the same storage location, since the value stored in that location will be zeroed out by the first XOR instruction, and then remain zero; it will not be "swapped with itself". This is not the same as if x and y have the same values. The trouble only comes when x and y use the same storage location, in which case their values must already be equal. That is, if x and y use the same storage location, then the line: sets x to zero (because x = y so X XOR Y is zero) and sets y to zero (since it uses the same storage location), causing x and y to lose their original values. == Proof of correctness == The binary operation XOR over bit strings of length N {\displaystyle N} exhibits the following properties (where ⊕ {\displaystyle \oplus } denotes XOR): L1. Commutativity: A ⊕ B = B ⊕ A {\displaystyle A\oplus B=B\oplus A} L2. Associativity: ( A ⊕ B ) ⊕ C = A ⊕ ( B ⊕ C ) {\displaystyle (A\oplus B)\oplus C=A\oplus (B\oplus C)} L3. Identity exists: there is a bit string, 0, (of length N) such that A ⊕ 0 = A {\displaystyle A\oplus 0=A} for any A {\displaystyle A} L4. Each element is its own inverse: for each A {\displaystyle A} , A ⊕ A = 0 {\displaystyle A\oplus A=0} . Suppose that we have two distinct registers R1 and R2 as in the table below, with initial values A and B respectively. We perform the operations below in sequence, and reduce our results using the properties listed above. === Linear algebra interpretation === As XOR can be interpreted as binary addition and a pair of bits can be interpreted as a vector in a two-dimensional vector space over the field with two elements, the steps in the algorithm can be interpreted as multiplication by 2×2 matrices over the field with two elements. For simplicity, assume initially that x and y are each single bits, not bit vectors. For example, the step: which also has the implicit: corresponds to the matrix ( 1 1 0 1 ) {\displaystyle \left({\begin{smallmatrix}1&1\\0&1\end{smallmatrix}}\right)} as ( 1 1 0 1 ) ( x y ) = ( x + y y ) . {\displaystyle {\begin{pmatrix}1&1\\0&1\end{pmatrix}}{\begin{pmatrix}x\\y\end{pmatrix}}={\begin{pmatrix}x+y\\y\end{pmatrix}}.} The sequence of operations is then expressed as: ( 1 1 0 1 ) ( 1 0 1 1 ) ( 1 1 0 1 ) = ( 0 1 1 0 ) {\displaystyle {\begin{pmatrix}1&1\\0&1\end{pmatrix}}{\begin{pmatrix}1&0\\1&1\end{pmatrix}}{\begin{pmatrix}1&1\\0&1\end{pmatrix}}={\begin{pmatrix}0&1\\1&0\end{pmatrix}}} (working with binary values, so 1 + 1 = 0 {\displaystyle 1+1=0} ), which expresses the elementary matrix of switching two rows (or columns) in terms of the transvections (shears) of adding one element to the other. To generalize to where X and Y are not single bits, but instead bit vectors of length n, these 2×2 matrices are replaced by 2n×2n block matrices such as ( I n I n 0 I n ) . {\displaystyle \left({\begin{smallmatrix}I_{n}&I_{n}\\0&I_{n}\end{smallmatrix}}\right).} These matrices are operating on values, not on variables (with storage locations), hence this interpretation abstracts away from issues of storage location and the problem of both variables sharing the same storage location. == Code example == A C function that implements the XOR swap algorithm: The code first checks if the addresses are distinct and uses a guard clause to exit the function early if they are equal. Without that check, if they were equal, the algorithm would fold to a triple x ^= x resulting in zero. == Reasons for avoidance in practice == On modern CPU architectures, the XOR technique can be slower than using a temporary variable to do swapping. At least on recent x86 CPUs, both by AMD and Intel, moving between registers regularly incurs zero latency. (This is called MOV-elimination.) Even if there is not any architectural register available to use, the XCHG instruction will be at least as fast as the three XORs taken together. Another reason is that modern CPUs strive to execute instructions in parallel via instruction pipelines. In the XOR technique, the inputs to each operation depend on the results of the previous operation, so they must be executed in strictly sequential order, negating any benefits of instruction-level parallelism. === Aliasing === The XOR swap is also complicated in practice by aliasing. If an attempt is made to XOR-swap the contents of some location with itself, the result is that the location is zeroed out and its value lost. Therefore, XOR swapping must not be used blindly in a high-level language if aliasing is possible. This issue does not apply if the technique is used in assembly to swap the contents of two registers. Similar problems occur with call by name, as in Jensen's Device, where swapping i and A[i] via a temporary variable yields incorrect results due to the arguments being related: swapping via temp = i; i = A[i]; A[i] = temp changes the value for i in the second statement, which then results in the incorrect i value for A[i] in the third statement. == Variations == The underlying principle of the XOR swap algorithm can be applied to any operation meeting criteria L1 through L4 above. Replacing XOR by addition and subtraction gives various slightly different, but largely equivalent, formulations. For example: Unlike the XOR swap, this variation requires that the underlying processor or programming language uses a method such as modular arithmetic or bignums to guarantee that the computation of X + Y cannot cause an error due to integer overflow. Therefore, it is seen even more rarely in practice than the XOR swap. However, the implementation of AddSwap above in the C programming language always works even in case of integer overflow, since, according to the C standard, addition and subtraction of unsigned integers follow the rules of modular arithmetic, i. e. are done in the cyclic group Z / 2 s Z {\displaystyle \mathbb {Z} /2^{s}\mathbb {Z} } where s {\displaystyle s} is the number of bits of unsigned int. Indeed, the correctness of the algorithm follows from the fact that the formulas ( x + y ) − y = x {\displaystyle (x+y)-y=x} and ( x + y ) − ( ( x + y ) − y ) = y {\displaystyle (x+y)-((x+y)-y)=y} hold in any abelian group. This generalizes the proof for the XOR swap algorithm: XOR is both the addition and subtraction in the abelian group ( Z / 2 Z ) s {\displaystyle (\mathbb {Z} /2\mathbb {Z} )^{s}} (which is the direct sum of s copies of Z / 2 Z {\displaystyle \mathbb {Z} /2\mathbb {Z} } ). This doesn't hold when dealing with the signed int type (the default for int). Signed integer overflow is an undefined behavior in C and thus modular arithmetic is not guaranteed by the standard, which may lead to incorrect results. The sequence of operations in AddSwap can be expressed via matrix multiplication as: ( 1 − 1 0 1 ) ( 1 0 1 − 1 ) ( 1 1 0 1 ) = ( 0 1 1 0 ) {\displaystyle {\begin{pmatrix}1&-1\\0&1\end{pmatrix}}{\begin{pmatrix}1&0\\1&-1\end{pmatrix}}{\begin{pmatrix}1&1\\0&1\end{pmatrix}}={\begin{pmatrix}0&1\\1&0\end{pmatrix}}} == Application to register allocation == On architectures lacking a dedicated swap instruction, because it avoids the extra temporary register, the XOR swap algorithm is required for optimal register allocatio
Gallery software
Gallery software is software that helps the user publish or share photos, pictures, videos or other digital media. Most galleries are located on Web servers, where users are allowed to register and publish their pictures. Gallery software usually features automatic image resizing, allows digital media be categorized into sets, and allows comments. == Types == Early digital media publishing and sharing was done with imageboards. The boards are by topics, sometimes called "chan". Each discussion in a "chan" are started with a piece of digital media, and follow-up discussions can contain another piece too. Software works in this way: Futallaby, Danbooru. Traditionally, galleries are managed. An administrator maintains a set of or hierarchy of albums. The users can upload their digital media in one of the existing albums defined by an administrator, or create their own albums. The users with sufficient permission can re-categorise the digital media others uploaded. Often, the site's administrator can define which album the users are allowed to categorise their media into, or delete other user's content. Examples are open source galleries Coppermine, Gallery Project. There are decentralised gallery software that does not have an administrator for managing contents. Pinterest, Flickr and DeviantArt has been successful with this model. Open source gallery software MediaGoblin works in this way. Each user can create their own "collections", to categorise theirs or other users' media. However users cannot put media into other user's collections. Each user's category is separate. There is no centralised theme or hierarchy for the media.
Deep learning in photoacoustic imaging
Photoacoustic imaging (PA) is based on the photoacoustic effect, in which optical absorption causes a rise in temperature, which causes a subsequent rise in pressure via thermo-elastic expansion. This pressure rise propagates through the tissue and is sensed via ultrasonic transducers. Due to the proportionality between the optical absorption, the rise in temperature, and the rise in pressure, the ultrasound pressure wave signal can be used to quantify the original optical energy deposition within the tissue. Photoacoustic imaging has applications of deep learning in both photoacoustic computed tomography (PACT) and photoacoustic microscopy (PAM). PACT utilizes wide-field optical excitation and an array of unfocused ultrasound transducers. Similar to other computed tomography methods, the sample is imaged at multiple view angles, which are then used to perform an inverse reconstruction algorithm based on the detection geometry (typically through universal backprojection, modified delay-and-sum, or time reversal ) to elicit the initial pressure distribution within the tissue. PAM on the other hand uses focused ultrasound detection combined with weakly focused optical excitation (acoustic resolution PAM or AR-PAM) or tightly focused optical excitation (optical resolution PAM or OR-PAM). PAM typically captures images point-by-point via a mechanical raster scanning pattern. At each scanned point, the acoustic time-of-flight provides axial resolution while the acoustic focusing yields lateral resolution. == Applications of deep learning in PACT == The first application of deep learning in PACT was by Reiter et al. in which a deep neural network was trained to learn spatial impulse responses and locate photoacoustic point sources. The resulting mean axial and lateral point location errors on 2,412 of their randomly selected test images were 0.28 mm and 0.37 mm respectively. After this initial implementation, the applications of deep learning in PACT have branched out primarily into removing artifacts from acoustic reflections, sparse sampling, limited-view, and limited-bandwidth. There has also been some recent work in PACT toward using deep learning for wavefront localization. There have been networks based on fusion of information from two different reconstructions to improve the reconstruction using deep learning fusion based networks. === Using deep learning to locate photoacoustic point sources === Traditional photoacoustic beamforming techniques modeled photoacoustic wave propagation by using detector array geometry and the time-of-flight to account for differences in the PA signal arrival time. However, this technique failed to account for reverberant acoustic signals caused by acoustic reflection, resulting in acoustic reflection artifacts that corrupt the true photoacoustic point source location information. In Reiter et al., a convolutional neural network (similar to a simple VGG-16 style architecture) was used that took pre-beamformed photoacoustic data as input and outputted a classification result specifying the 2-D point source location. ==== Deep learning for PA wavefront localization ==== Johnstonbaugh et al. was able to localize the source of photoacoustic wavefronts with a deep neural network. The network used was an encoder-decoder style convolutional neural network. The encoder-decoder network was made of residual convolution, upsampling, and high field-of-view convolution modules. A Nyquist convolution layer and differentiable spatial-to-numerical transform layer were also used within the architecture. Simulated PA wavefronts served as the input for training the model. To create the wavefronts, the forward simulation of light propagation was done with the NIRFast toolbox and the light-diffusion approximation, while the forward simulation of sound propagation was done with the K-Wave toolbox. The simulated wavefronts were subjected to different scattering mediums and Gaussian noise. The output for the network was an artifact free heat map of the targets axial and lateral position. The network had a mean error rate of less than 30 microns when localizing target below 40 mm and had a mean error rate of 1.06 mm for localizing targets between 40 mm and 60 mm. With a slight modification to the network, the model was able to accommodate multi target localization. A validation experiment was performed in which pencil lead was submerged into an intralipid solution at a depth of 32 mm. The network was able to localize the lead's position when the solution had a reduced scattering coefficient of 0, 5, 10, and 15 cm−1. The results of the network show improvements over standard delay-and-sum or frequency-domain beamforming algorithms and Johnstonbaugh proposes that this technology could be used for optical wavefront shaping, circulating melanoma cell detection, and real-time vascular surgeries. === Removing acoustic reflection artifacts (in the presence of multiple sources and channel noise) === Building on the work of Reiter et al., Allman et al. utilized a full VGG-16 architecture to locate point sources and remove reflection artifacts within raw photoacoustic channel data (in the presence of multiple sources and channel noise). This utilization of deep learning trained on simulated data produced in the MATLAB k-wave library, and then later reaffirmed their results on experimental data. === Ill-posed PACT reconstruction === In PACT, tomographic reconstruction is performed, in which the projections from multiple solid angles are combined to form an image. When reconstruction methods like filtered backprojection or time reversal, are ill-posed inverse problems due to sampling under the Nyquist-Shannon's sampling requirement or with limited-bandwidth/view, the resulting reconstruction contains image artifacts. Traditionally these artifacts were removed with slow iterative methods like total variation minimization, but the advent of deep learning approaches has opened a new avenue that utilizes a priori knowledge from network training to remove artifacts. In the deep learning methods that seek to remove these sparse sampling, limited-bandwidth, and limited-view artifacts, the typical workflow involves first performing the ill-posed reconstruction technique to transform the pre-beamformed data into a 2-D representation of the initial pressure distribution that contains artifacts. Then, a convolutional neural network (CNN) is trained to remove the artifacts, in order to produce an artifact-free representation of the ground truth initial pressure distribution. ==== Using deep learning to remove sparse sampling artifacts ==== When the density of uniform tomographic view angles is under what is prescribed by the Nyquist-Shannon's sampling theorem, it is said that the imaging system is performing sparse sampling. Sparse sampling typically occurs as a way of keeping production costs low and improving image acquisition speed. The typical network architectures used to remove these sparse sampling artifacts are U-net and Fully Dense (FD) U-net. Both of these architectures contain a compression and decompression phase. The compression phase learns to compress the image to a latent representation that lacks the imaging artifacts and other details. The decompression phase then combines with information passed by the residual connections in order to add back image details without adding in the details associated with the artifacts. FD U-net modifies the original U-net architecture by including dense blocks that allow layers to utilize information learned by previous layers within the dense block. Another technique was proposed using a simple CNN based architecture for removal of artifacts and improving the k-wave image reconstruction. ==== Removing limited-view artifacts with deep learning ==== When a region of partial solid angles are not captured, generally due to geometric limitations, the image acquisition is said to have limited-view. As illustrated by the experiments of Davoudi et al., limited-view corruptions can be directly observed as missing information in the frequency domain of the reconstructed image. Limited-view, similar to sparse sampling, makes the initial reconstruction algorithm ill-posed. Prior to deep learning, the limited-view problem was addressed with complex hardware such as acoustic deflectors and full ring-shaped transducer arrays, as well as solutions like compressed sensing, weighted factor, and iterative filtered backprojection. The result of this ill-posed reconstruction is imaging artifacts that can be removed by CNNs. The deep learning algorithms used to remove limited-view artifacts include U-net and FD U-net, as well as generative adversarial networks (GANs) and volumetric versions of U-net. One GAN implementation of note improved upon U-net by using U-net as a generator and VGG as a discriminator, with the Wasserstein metric and gradient penalty to stabilize training (WGAN-GP). ==== Pixel-wise interpolation
Pooling layer
In neural networks, a pooling layer is a kind of network layer that downsamples and aggregates information that is dispersed among many vectors into fewer vectors. It has several uses. It removes redundant information, thus reducing the amount of computation and memory required, which makes the model more robust to small variations in the input; and it increases the receptive field of neurons in later layers in the network. == Convolutional neural network pooling == Pooling is most commonly used in convolutional neural networks (CNN). Below is a description of pooling in 2-dimensional CNNs. The generalization to n-dimensions is immediate. As notation, we consider a tensor x ∈ R H × W × C {\displaystyle x\in \mathbb {R} ^{H\times W\times C}} , where H {\displaystyle H} is height, W {\displaystyle W} is width, and C {\displaystyle C} is the number of channels. A pooling layer outputs a tensor y ∈ R H ′ × W ′ × C ′ {\displaystyle y\in \mathbb {R} ^{H'\times W'\times C'}} . We define two variables f , s {\displaystyle f,s} called "filter size" (aka "kernel size") and "stride". Sometimes, it is necessary to use a different filter size and stride for horizontal and vertical directions. In such cases, we define 4 variables: f H , f W , s H , s W {\displaystyle f_{H},f_{W},s_{H},s_{W}} . The receptive field of an entry in the output tensor, y {\displaystyle y} , are all the entries in x {\displaystyle x} that can affect that entry. === Max pooling === Max Pooling (MaxPool) is commonly used in CNNs to reduce the spatial dimensions of feature maps. Define M a x P o o l ( x | f , s ) 0 , 0 , 0 = max ( x 0 : f − 1 , 0 : f − 1 , 0 ) {\displaystyle \mathrm {MaxPool} (x|f,s)_{0,0,0}=\max(x_{0:f-1,0:f-1,0})} where 0 : f − 1 {\displaystyle 0:f-1} means the range 0 , 1 , … , f − 1 {\displaystyle 0,1,\dots ,f-1} . Note that we need to avoid the off-by-one error. The next input is M a x P o o l ( x | f , s ) 1 , 0 , 0 = max ( x s : s + f − 1 , 0 : f − 1 , 0 ) {\displaystyle \mathrm {MaxPool} (x|f,s)_{1,0,0}=\max(x_{s:s+f-1,0:f-1,0})} and so on. The receptive field of y i , j , c {\displaystyle y_{i,j,c}} is x i s + f − 1 , j s + f − 1 , c {\displaystyle x_{is+f-1,js+f-1,c}} , so in general, M a x P o o l ( x | f , s ) i , j , c = m a x ( x i s : i s + f − 1 , j s : j s + f − 1 , c ) {\displaystyle \mathrm {MaxPool} (x|f,s)_{i,j,c}=\mathrm {max} (x_{is:is+f-1,js:js+f-1,c})} If the horizontal and vertical filter size and strides differ, then in general, M a x P o o l ( x | f , s ) i , j , c = m a x ( x i s H : i s H + f H − 1 , j s W : j s W + f W − 1 , c ) {\displaystyle \mathrm {MaxPool} (x|f,s)_{i,j,c}=\mathrm {max} (x_{is_{H}:is_{H}+f_{H}-1,js_{W}:js_{W}+f_{W}-1,c})} More succinctly, we can write y k = max ( { x k ′ | k ′ in the receptive field of k } ) {\displaystyle y_{k}=\max(\{x_{k'}|k'{\text{ in the receptive field of }}k\})} . If H {\displaystyle H} is not expressible as k s + f {\displaystyle ks+f} where k {\displaystyle k} is an integer, then for computing the entries of the output tensor on the boundaries, max pooling would attempt to take as inputs variables off the tensor. In this case, how those non-existent variables are handled depends on the padding conditions, illustrated on the right. Global Max Pooling (GMP) is a specific kind of max pooling where the output tensor has shape R C {\displaystyle \mathbb {R} ^{C}} and the receptive field of y c {\displaystyle y_{c}} is all of x 0 : H , 0 : W , c {\displaystyle x_{0:H,0:W,c}} . That is, it takes the maximum over each entire channel. It is often used just before the final fully connected layers in a CNN classification head. === Average pooling === Average pooling (AvgPool) is similarly defined A v g P o o l ( x | f , s ) i , j , c = a v e r a g e ( x i s : i s + f − 1 , j s : j s + f − 1 , c ) = 1 f 2 ∑ k ∈ i s : i s + f − 1 ∑ l ∈ j s : j s + f − 1 x k , l , c {\displaystyle \mathrm {AvgPool} (x|f,s)_{i,j,c}=\mathrm {average} (x_{is:is+f-1,js:js+f-1,c})={\frac {1}{f^{2}}}\sum _{k\in is:is+f-1}\sum _{l\in js:js+f-1}x_{k,l,c}} Global Average Pooling (GAP) is defined similarly to GMP. It was first proposed in Network-in-Network. Similarly to GMP, it is often used just before the final fully connected layers in a CNN classification head. === Interpolations === There are some interpolations of max pooling and average pooling. Mixed Pooling is a linear sum of max pooling and average pooling. That is, M i x e d P o o l ( x | f , s , w ) = w M a x P o o l ( x | f , s ) + ( 1 − w ) A v g P o o l ( x | f , s ) {\displaystyle \mathrm {MixedPool} (x|f,s,w)=w\mathrm {MaxPool} (x|f,s)+(1-w)\mathrm {AvgPool} (x|f,s)} where w ∈ [ 0 , 1 ] {\displaystyle w\in [0,1]} is either a hyperparameter, a learnable parameter, or randomly sampled anew every time. Lp Pooling is similar to average pooling, but uses Lp norm average instead of average: y k = ( 1 N ∑ k ′ in the receptive field of k | x k ′ | p ) 1 / p {\displaystyle y_{k}=\left({\frac {1}{N}}\sum _{k'{\text{ in the receptive field of }}k}|x_{k'}|^{p}\right)^{1/p}} where N {\displaystyle N} is the size of receptive field, and p ≥ 1 {\displaystyle p\geq 1} is a hyperparameter. If all activations are non-negative, then average pooling is the case of p = 1 {\displaystyle p=1} , and max pooling is the case of p → ∞ {\displaystyle p\to \infty } . Square-root pooling is the case of p = 2 {\displaystyle p=2} . Stochastic pooling samples a random activation x k ′ {\displaystyle x_{k'}} from the receptive field with probability x k ′ ∑ k ″ x k ″ {\displaystyle {\frac {x_{k'}}{\sum _{k''}x_{k''}}}} . It is the same as average pooling in expectation. Softmax pooling is like max pooling, but uses softmax, i.e. ∑ k ′ e β x k ′ x k ′ ∑ k ″ e β x k ″ {\displaystyle {\frac {\sum _{k'}e^{\beta x_{k'}}x_{k'}}{\sum _{k''}e^{\beta x_{k''}}}}} where β > 0 {\displaystyle \beta >0} . Average pooling is the case of β ↓ 0 {\displaystyle \beta \downarrow 0} , and max pooling is the case of β ↑ ∞ {\displaystyle \beta \uparrow \infty } Local Importance-based Pooling generalizes softmax pooling by ∑ k ′ e g ( x k ′ ) x k ′ ∑ k ″ e g ( x k ″ ) {\displaystyle {\frac {\sum _{k'}e^{g(x_{k'})}x_{k'}}{\sum _{k''}e^{g(x_{k''})}}}} where g {\displaystyle g} is a learnable function. === Other poolings === Spatial pyramidal pooling applies max pooling (or any other form of pooling) in a pyramid structure. That is, it applies global max pooling, then applies max pooling to the image divided into 4 equal parts, then 16, etc. The results are then concatenated. It is a hierarchical form of global pooling, and similar to global pooling, it is often used just before a classification head. Region of Interest Pooling (also known as RoI pooling) is a variant of max pooling used in R-CNNs for object detection. It is designed to take an arbitrarily-sized input matrix, and output a fixed-sized output matrix. Covariance pooling computes the covariance matrix of the vectors { x k , l , 0 : C − 1 } k ∈ i s : i s + f − 1 , l ∈ j s : j s + f − 1 {\displaystyle \{x_{k,l,0:C-1}\}_{k\in is:is+f-1,l\in js:js+f-1}} which is then flattened to a C 2 {\displaystyle C^{2}} -dimensional vector y i , j , 0 : C 2 − 1 {\displaystyle y_{i,j,0:C^{2}-1}} . Global covariance pooling is used similarly to global max pooling. As average pooling computes the average, which is a first-degree statistic, and covariance is a second-degree statistic, covariance pooling is also called "second-order pooling". It can be generalized to higher-order poolings. Blur Pooling means applying a blurring method before downsampling. For example, the Rect-2 blur pooling means taking an average pooling at f = 2 , s = 1 {\displaystyle f=2,s=1} , then taking every second pixel (identity with s = 2 {\displaystyle s=2} ). == Vision Transformer pooling == In Vision Transformers (ViT), there are the following common kinds of poolings. BERT-like pooling uses a dummy [CLS] token, "classification". For classification, the output at [CLS] is the classification token, which is then processed by a LayerNorm-feedforward-softmax module into a probability distribution, which is the network's prediction of class probability distribution. This is the one used by the original ViT and Masked Autoencoder. Global average pooling (GAP) does not use the dummy token, but simply takes the average of all output tokens as the classification token. It was mentioned in the original ViT as being equally good. Multihead attention pooling (MAP) applies a multi headed attention block to pooling. Specifically, it takes as input a list of vectors x 1 , x 2 , … , x n {\displaystyle x_{1},x_{2},\dots ,x_{n}} , which might be thought of as the output vectors of a layer of a ViT. It then applies a feedforward layer F F N {\displaystyle \mathrm {FFN} } on each vector, resulting in a matrix V = [ F F N ( v 1 ) , … , F F N ( v n ) ] {\displaystyle V=[\mathrm {FFN} (v_{1}),\dots ,\mathrm {FFN} (v_{n})]} . This is then sent to a multi-headed attention, resulting in M u l t i h e a d e d A
PropBank
PropBank is a corpus that is annotated with verbal propositions and their arguments—a "proposition bank". Although "PropBank" refers to a specific corpus produced by Martha Palmer et al., the term propbank is also coming to be used as a common noun referring to any corpus that has been annotated with propositions and their arguments. The PropBank project has played a role in research in natural language processing, and has been used in semantic role labelling. == Comparison == PropBank differs from FrameNet, the resource to which it is most frequently compared, in several ways. PropBank is a verb-oriented resource, while FrameNet is centered on the more abstract notion of frames, which generalizes descriptions across similar verbs (e.g. "describe" and "characterize") as well as nouns and other words (e.g. "description"). PropBank does not annotate events or states of affairs described using nouns. PropBank commits to annotating all verbs in a corpus, whereas the FrameNet project chooses sets of example sentences from a large corpus and only in a few cases has annotated longer continuous stretches of text. PropBank-style annotations often remain close to the syntactic level, while FrameNet-style annotations are sometimes more semantically motivated. From the start, PropBank was developed with the idea of serving as training data for machine learning-based semantic role labeling systems in mind. It requires that all arguments to a verb be syntactic constituents and different senses of a word are only distinguished if the differences bear on the arguments. Due to such differences, semantic role labeling with respect to PropBank is often a somewhat easier task than producing FrameNet-style annotations.
KoalaPad
The KoalaPad is a graphics tablet, released in 1983 by US company Koala Technologies Corporation, for the Apple II, TRS-80 Color Computer (as the TRS-80 Touch Pad), Atari 8-bit computers, Commodore 64, and IBM PC compatibles. Originally designed by Dr. David Thornburg as a low-cost computer drawing tool for schools, the Koala Pad and the bundled drawing program, KoalaPainter, was popular with home users as well. KoalaPainter was called KoalaPaint in some versions for the Apple II, and PC Design for the IBM PC. A program called Graphics Exhibitor was included for creating slideshow presentations from KoalaPainter drawings. == Description == The pad was four inches square (i.e. roughly 10×10 cm) and mounted on a slightly inclined base with the back of the pad higher than the front. At the top, "behind" the pad, were two buttons. The pad hooked into the computer using the analog signals of the joystick ports (the so-called paddle inputs), which meant that it had a low resolution and tended to jostle the cursor if moved during use. As an alternative to the drawing stylus, the pad could as easily be operated by the user's fingers for tasks that demanded less precision, such as selecting between menu items (thus using the pad as a kind of "indirect touch screen"). The top-mounted buttons tended to be somewhat frustrating to use, as the user had to "reach around" the stylus to push the buttons in order to start or stop drawing. A similar tablet from Atari, the Atari CX77 Touch Tablet, addressed this with a built-in button on the stylus, which some enterprising users adapted for use with their KoalaPad. == KoalaPainter == The pad shipped with a simple bitmap graphics editor developed by Audio Light called KoalaPainter, PC Design or Micro Illustrator depending on the target machine (see release history). Although bundled with the pad, KoalaPainter could also be operated using an ordinary digital joystick. One unique feature of the program, for its time, was that it held two pictures in the computer's memory, allowing the user to flip from one to the other—a function commonly used in order to study the differences between an original and a modified picture, and to copy and paste between two different pictures. Some third-party bitmap editors could also be used with the KoalaPad, such as Broderbund's Dazzle Draw for the Apple II. === Release history === KoalaPainter for Commodore 64 (1983) and Atari 8-bit computers (1983) PC Design for the IBM PC (1983) Micro Illustrator for the Apple II (1983), Atari 8-bit computers (1983) and Commodore Plus/4 (1984) KoalaPainter II for Commodore 64 (1984) === Reception === Ahoy! called KoalaPainter "a very powerful and effective color drawing package", and concluded that it and the KoalaPad were "excellent in ease of use, a fine choice for a beginner as well as young children". BYTE's reviewer stated in December 1984 that he made far fewer errors when using an Apple Mouse with MousePaint than with a KoalaPad and its software. He found that MousePaint was easier to use and more efficient, predicting that the mouse would receive more software support than the pad. Cassie Stahl in InfoWorld's Essential Guide to Atari Computers praised the tablet and its documentation, rating it "Excellent" among all categories and stating that "Playing with the KoalaPad becomes addictive. It does everything it claims to, and it does it well". She also liked Micro Illustrator, rating it "Excellent" except for "Good" for Performance. While criticizing the limited erase function, Stahl reported an undocumented feature enabling exporting pictures to other software. === File format === The Commodore 64 version of KoalaPainter used a fairly simple file format corresponding directly to the way bitmapped graphics are handled on the computer: A two-byte load address, followed immediately by 8,000 bytes of raw bitmap data, 1,000 bytes of raw "Video Matrix" data, 1,000 bytes of raw "Color RAM" data, and a one-byte Background Color field. == KoalaWare == Koala Technologies offered more software beyond the bundled KoalaPainter and Graphics Exhibitor for use with the pad. Among these applications, marketed under the moniker KoalaWare (like KoalaPainter itself), was educational software for use with customized keypads and overlays, such as spelling tools, music programs, and mathematics instruction software, as well as software for "translating" graphical designs into Logo programs.
Kdan Mobile
Kdan Mobile Software Limited is a software application development company based in Tainan City, Taiwan. Kdan also has branches in Taipei, Changsha, Irvine, California, Japan, and South Korea. The company was founded in 2009 by Kenny Su, the company's CEO. == History == Kdan Mobile was founded in 2009 by Kenny Su (蘇柏州) and develops an application for PDF documents. Su previously worked at the Industrial Technology Research Institute (ITRI) . In 2018, the company completed its Series B round of fundraising, in which it raised 16 million USD in total. Four global firms, Dattoz Partners (South Korea), WI Harper Group (U.S.), Taiwania Capital (Taiwan), and Golden Asia Fund Mitsubishi UFJ Capital (Japan), made up the Series B investment. Kdan previously raised 5 million USD in its Series A round in 2018.