In electrical engineering, statistical computing and bioinformatics, the Baum–Welch algorithm is a special case of the expectation–maximization algorithm used to find the unknown parameters of a hidden Markov model (HMM). It makes use of the forward-backward algorithm to compute the statistics for the expectation step. The Baum–Welch algorithm, the primary method for inference in hidden Markov models, is numerically unstable due to its recursive calculation of joint probabilities. As the number of variables grows, these joint probabilities become increasingly small, leading to the forward recursions rapidly approaching values below machine precision. == History == The Baum–Welch algorithm was named after its inventors Leonard E. Baum and Lloyd R. Welch. The algorithm and the Hidden Markov models were first described in a series of articles by Baum and his peers at the IDA Center for Communications Research, Princeton in the late 1960s and early 1970s. One of the first major applications of HMMs was to the field of speech processing. In the 1980s, HMMs were emerging as a useful tool in the analysis of biological systems and information, and in particular genetic information. They have since become an important tool in the probabilistic modeling of genomic sequences. == Description == A hidden Markov model describes the joint probability of a collection of "hidden" and observed discrete random variables. It relies on the assumption that the i-th hidden variable given the (i − 1)-th hidden variable is independent of previous hidden variables, and the current observation variables depend only on the current hidden state. The Baum–Welch algorithm uses the well known EM algorithm to find the maximum likelihood estimate of the parameters of a hidden Markov model given a set of observed feature vectors. Let X t {\displaystyle X_{t}} be a discrete hidden random variable with N {\displaystyle N} possible values (i.e. We assume there are N {\displaystyle N} states in total). We assume the P ( X t ∣ X t − 1 ) {\displaystyle P(X_{t}\mid X_{t-1})} is independent of time t {\displaystyle t} , which leads to the definition of the time-independent stochastic transition matrix A = { a i j } = P ( X t = j ∣ X t − 1 = i ) . {\displaystyle A=\{a_{ij}\}=P(X_{t}=j\mid X_{t-1}=i).} The initial state distribution (i.e. when t = 1 {\displaystyle t=1} ) is given by π i = P ( X 1 = i ) . {\displaystyle \pi _{i}=P(X_{1}=i).} The observation variables Y t {\displaystyle Y_{t}} can take one of K {\displaystyle K} possible values. We also assume the observation given the "hidden" state is time independent. The probability of a certain observation y i {\displaystyle y_{i}} at time t {\displaystyle t} for state X t = j {\displaystyle X_{t}=j} is given by b j ( y i ) = P ( Y t = y i ∣ X t = j ) . {\displaystyle b_{j}(y_{i})=P(Y_{t}=y_{i}\mid X_{t}=j).} Taking into account all the possible values of Y t {\displaystyle Y_{t}} and X t {\displaystyle X_{t}} , we obtain the N × K {\displaystyle N\times K} matrix B = { b j ( y i ) } {\displaystyle B=\{b_{j}(y_{i})\}} where b j {\displaystyle b_{j}} belongs to all the possible states and y i {\displaystyle y_{i}} belongs to all the observations. An observation sequence is given by Y = ( Y 1 = y 1 , Y 2 = y 2 , … , Y T = y T ) {\displaystyle Y=(Y_{1}=y_{1},Y_{2}=y_{2},\ldots ,Y_{T}=y_{T})} . Thus we can describe a hidden Markov chain by θ = ( A , B , π ) {\displaystyle \theta =(A,B,\pi )} . The Baum–Welch algorithm finds a local maximum for θ ∗ = a r g m a x θ P ( Y ∣ θ ) {\displaystyle \theta ^{}=\operatorname {arg\,max} _{\theta }P(Y\mid \theta )} (i.e. the HMM parameters θ {\displaystyle \theta } that maximize the probability of the observation). === Algorithm === Set θ = ( A , B , π ) {\displaystyle \theta =(A,B,\pi )} with random initial conditions. They can also be set using prior information about the parameters if it is available; this can speed up the algorithm and also steer it toward the desired local maximum. ==== Forward procedure ==== Let α i ( t ) = P ( Y 1 = y 1 , … , Y t = y t , X t = i ∣ θ ) {\displaystyle \alpha _{i}(t)=P(Y_{1}=y_{1},\ldots ,Y_{t}=y_{t},X_{t}=i\mid \theta )} , the probability of seeing the observations y 1 , y 2 , … , y t {\displaystyle y_{1},y_{2},\ldots ,y_{t}} and being in state i {\displaystyle i} at time t {\displaystyle t} . This is found recursively: α i ( 1 ) = π i b i ( y 1 ) , {\displaystyle \alpha _{i}(1)=\pi _{i}b_{i}(y_{1}),} α i ( t + 1 ) = b i ( y t + 1 ) ∑ j = 1 N α j ( t ) a j i . {\displaystyle \alpha _{i}(t+1)=b_{i}(y_{t+1})\sum _{j=1}^{N}\alpha _{j}(t)a_{ji}.} Since this series converges exponentially to zero, the algorithm will numerically underflow for longer sequences. However, this can be avoided in a slightly modified algorithm by scaling α {\displaystyle \alpha } in the forward and β {\displaystyle \beta } in the backward procedure below. ==== Backward procedure ==== Let β i ( t ) = P ( Y t + 1 = y t + 1 , … , Y T = y T ∣ X t = i , θ ) {\displaystyle \beta _{i}(t)=P(Y_{t+1}=y_{t+1},\ldots ,Y_{T}=y_{T}\mid X_{t}=i,\theta )} that is the probability of the ending partial sequence y t + 1 , … , y T {\displaystyle y_{t+1},\ldots ,y_{T}} given starting state i {\displaystyle i} at time t {\displaystyle t} . We calculate β i ( t ) {\displaystyle \beta _{i}(t)} as, β i ( T ) = 1 , {\displaystyle \beta _{i}(T)=1,} β i ( t ) = ∑ j = 1 N β j ( t + 1 ) a i j b j ( y t + 1 ) . {\displaystyle \beta _{i}(t)=\sum _{j=1}^{N}\beta _{j}(t+1)a_{ij}b_{j}(y_{t+1}).} ==== Update ==== We can now calculate the temporary variables, according to Bayes' theorem: γ i ( t ) = P ( X t = i ∣ Y , θ ) = P ( X t = i , Y ∣ θ ) P ( Y ∣ θ ) = α i ( t ) β i ( t ) ∑ j = 1 N α j ( t ) β j ( t ) , {\displaystyle \gamma _{i}(t)=P(X_{t}=i\mid Y,\theta )={\frac {P(X_{t}=i,Y\mid \theta )}{P(Y\mid \theta )}}={\frac {\alpha _{i}(t)\beta _{i}(t)}{\sum _{j=1}^{N}\alpha _{j}(t)\beta _{j}(t)}},} which is the probability of being in state i {\displaystyle i} at time t {\displaystyle t} given the observed sequence Y {\displaystyle Y} and the parameters θ {\displaystyle \theta } ξ i j ( t ) = P ( X t = i , X t + 1 = j ∣ Y , θ ) = P ( X t = i , X t + 1 = j , Y ∣ θ ) P ( Y ∣ θ ) = α i ( t ) a i j β j ( t + 1 ) b j ( y t + 1 ) ∑ k = 1 N ∑ w = 1 N α k ( t ) a k w β w ( t + 1 ) b w ( y t + 1 ) , {\displaystyle \xi _{ij}(t)=P(X_{t}=i,X_{t+1}=j\mid Y,\theta )={\frac {P(X_{t}=i,X_{t+1}=j,Y\mid \theta )}{P(Y\mid \theta )}}={\frac {\alpha _{i}(t)a_{ij}\beta _{j}(t+1)b_{j}(y_{t+1})}{\sum _{k=1}^{N}\sum _{w=1}^{N}\alpha _{k}(t)a_{kw}\beta _{w}(t+1)b_{w}(y_{t+1})}},} which is the probability of being in state i {\displaystyle i} and j {\displaystyle j} at times t {\displaystyle t} and t + 1 {\displaystyle t+1} respectively given the observed sequence Y {\displaystyle Y} and parameters θ {\displaystyle \theta } . The denominators of γ i ( t ) {\displaystyle \gamma _{i}(t)} and ξ i j ( t ) {\displaystyle \xi _{ij}(t)} are the same ; they represent the probability of making the observation Y {\displaystyle Y} given the parameters θ {\displaystyle \theta } . The parameters of the hidden Markov model θ {\displaystyle \theta } can now be updated: π i ∗ = γ i ( 1 ) , {\displaystyle \pi _{i}^{}=\gamma _{i}(1),} which is the expected frequency spent in state i {\displaystyle i} at time 1 {\displaystyle 1} . a i j ∗ = ∑ t = 1 T − 1 ξ i j ( t ) ∑ t = 1 T − 1 γ i ( t ) , {\displaystyle a_{ij}^{}={\frac {\sum _{t=1}^{T-1}\xi _{ij}(t)}{\sum _{t=1}^{T-1}\gamma _{i}(t)}},} which is the expected number of transitions from state i to state j compared to the expected total number of transitions starting in state i, including from state i to itself. The number of transitions starting in state i is equivalent to the number of times state i is observed in the sequence from t = 1 to t = T − 1. b i ∗ ( v k ) = ∑ t = 1 T 1 y t = v k γ i ( t ) ∑ t = 1 T γ i ( t ) , {\displaystyle b_{i}^{}(v_{k})={\frac {\sum _{t=1}^{T}1_{y_{t}=v_{k}}\gamma _{i}(t)}{\sum _{t=1}^{T}\gamma _{i}(t)}},} where 1 y t = v k = { 1 if y t = v k , 0 otherwise {\displaystyle 1_{y_{t}=v_{k}}={\begin{cases}1&{\text{if }}y_{t}=v_{k},\\0&{\text{otherwise}}\end{cases}}} is an indicator function, and b i ∗ ( v k ) {\displaystyle b_{i}^{}(v_{k})} is the expected number of times the output observations have been equal to v k {\displaystyle v_{k}} while in state i {\displaystyle i} over the expected total number of times in state i {\displaystyle i} . These steps are now repeated iteratively until a desired level of convergence. Note: It is possible to over-fit a particular data set. That is, P ( Y ∣ θ final ) > P ( Y ∣ θ true ) {\displaystyle P(Y\mid \theta _{\text{final}})>P(Y\mid \theta _{\text{true}})} . The algorithm also does not guarantee a global maximum. ==== Multiple sequences ==== The algorithm described thus far assumes a single observed sequence Y = y 1 , … , y T {\displaystyle Y=y_{1},\ldots ,y_{T}} . However, in many situations, there are several sequences observed: Y 1 ,
Diia
Diia (Ukrainian: Дія [ˈd⁽ʲ⁾ijɐ] , lit. 'Action'; also an acronym for Держава і Я, Derzhava i Ya, IPA: [derˈʒɑwɐ i ˈjɑ], lit. 'State and Me') is a mobile app, a web portal and a brand of e-governance in Ukraine. Launched in 2020, the Diia app allows Ukrainian citizens to use digital documents on their smartphones instead of physical ones for identification and sharing purposes. The Diia portal allows access to over 130 government services. Eventually, the government plans to make all kinds of state-person interactions available through Diia. Diia was built in partnership with the United States and is poised to be shared with other countries. On the sidelines of the 2023 World Economic Forum in Davos, USAID Administrator Samantha Power said the US hopes to replicate the success of Diia in other countries. == History == Diia was first presented on September 27, 2019, by the Ministry of Digital Transformation of Ukraine as a brand of the State in a Smartphone project. Vice Prime Minister and Minister of Digital Transformation Mykhailo Fedorov announced the creation of a mobile app and a web portal that would unite in a single place all the services provided by the state to citizens and businesses. On February 6, 2020, the mobile app Diia was officially launched. During the presentation, Ukrainian President Volodymyr Zelensky said that 9 million Ukrainians now have access to their driver's license and car registration documents on their phones, while Prime Minister Oleksiy Honcharuk called the implementation of the State in a Smartphone project a priority for the government. In April 2020, the Ukrainian government approved a resolution for experimental usage of digital ID-cards and passports which would be issued to all Ukrainians via the Diia. On October 5, 2020, during the Diia Summit, the government presented a first major update of the app and web portal branded "Diia 2.0". More types of documents were added to the app as well as the ability to share documents with others via a single tap on a push-message. The web portal in turn expanded the number of available services to 27, including the ability to register a private limited company in half an hour. President Zelensky who opened the summit, announced that in 2021 Ukraine will enter the "paper less" mode by prohibiting civil servants from demanding paper documents. By the end of 2020, the app had more than six million users, while the portal had 50 available services. In March 2021, the Ukrainian parliament adopted a bill equating digital identity documents with their physical analogues. Starting on August 23, Ukrainian citizens can use digital ID-cards and passports for all purposes while in Ukraine. According to Minister of Digital Transformation Mykhailo Fedorov, Ukraine will become the first country in the world where digital identity documents are considered legally equivalent to ordinary ones. In September 2024, Diia launched an online marriage registration service, which can be beneficial especially for military personnel who spend much time on the frontline separated from their partners. In October 2024, Diia's online marriage service appeared in Time's Inventions of the 2024 list. In the first month of its operations over 1.1 million Ukrainians tried to make proposals using the technology, and 435 couples got married. == Benefits and challenges == The first and most obvious benefit is the convenience of such a platform. Citizens can have many documents on their smartphones at once, without concern about losing or damaging them. Whenever needed, they can just open an app on their smartphones and show/check the document they need. The idea is that Diia will help cut the bureaucracy associated with public services, which in turn will help fight corruption and increase government savings. Fewer people are needed to be employed in the public sector and fewer human to human interactions are supposed to happen. With the start of the program, already 10% of government employees were reduced, which contributes to hundreds of millions of dollars in savings, but besides this, the initiative also improves the speed, efficiency, and transparency of government services. In addition, the digitalization of the government sector helps to develop the whole IT industry in the country, people become more digitally aware and educated, this affects other sectors as well, increasing the spread of digital infrastructure and expediting the speed of overall digitalization. The UN E-government Development Index, which assesses the capabilities of governments to integrate its functions electronically, such as the use of internet and mobile devices, ranked Ukraine 69th in 193 countries surveyed in 2020. Despite its low ranking in the e-government development index, Ukraine made a big jump on the e-participation index, which they ranked 43rd out of 193 countries from 0.66 in 2018 to 0.81 in 2020 (un.org, 2020), suggesting that the government and its citizens are adapting the IT-based government functions. The main goal of e-government according to Perez-Morote et.al. (2020) is to have accountability and transparency among the countries involved. But to do so, there are several challenges that a country should assess first prior to implementing e-government. In the research written by Heeks (2001), the author identified 2 main challenges that countries face in the development of e-government, first is the strategic challenge which involves the preparedness (e-readiness) of the entire government system for electronic transformation, and second challenge is the tactical challenge where the government must design (e-governance design) a system where it can be understood by every user, it's important that the information that needs to be communicated to the consumers is received clearly. For the first challenge (e-readiness), Ukraine had an internet penetration rate of 76% in 2020 and is expected to grow to 82%, it is important that consumers have the internet access for it to enable the consumers to utilize the service. Another factor is the readiness of its institutional infrastructure, which means that the government has its own organization which is solely focused on implementing the e-government project. In the case of Ukraine, the e-governance team is led by Oleksandr Ryzhenko, and the country's e-governance initiative is even further strengthened by ensuring that the data and legal infrastructure are already prepared. Ukraine has done this by modernizing their legislation that is more appropriate in the digital service, and the data exchange solution used by Ukraine is called Trembita. The human infrastructure is also being updated, as competent individuals must be the one doing the task, hence, EGOV4UKRAINE was launched, this aims to get IT developers for developing a system for administrative services. These efforts by the Ukrainian government did not go unnoticed, and they received an award from the e-Governance Academy as "partner of the year 2017". For the second challenge, which deals with the system design, the success of Ukraine can be seen on the latest data of UNDP, where it shows a high increase in the E-participation index. In 2018, Ukraine ranked 75th it ranked 46th in 2020 (un.org, 2020). Despite visible success, the implementation of the e-government was accompanied by problems. Data leakage became the main one. In May 2020, the data of 26 million driver's licenses appeared in the public domain on the Internet. The Ukrainian government said the Diia app was not linked to a data breach, but it is impossible to say for certain. Any storage of official documents in electronic format is associated with the risk of their leakage. In addition, the Diia application still has data protection issues, as the required protection system has not been implemented. This is also compounded by the country's weak data protection legal regime. In addition, since 2023, Ukrainians are able to register their cars with this app. Issued license plates are not using regional codes, but they are using special codes starting with DI or PD. == Diia City == In May 2020, the government presented Diia City headed by Oleksandr Borniakov, a large-scale project which would establish a virtual model of a free economic zone for representatives of the creative economy. It would provide for special digital residency with a particular taxation regime, intellectual property protection and simplified regulations. Diia City concurrently imposes certain constraints on contracts involving individual entrepreneurs (FOPs). It also offers the benefit of tax rebates. Diia City garners endorsement from the Ukrainian government, believing it will support the country's position in the IT market. As of July 30, 2023, the program had more than 600 residents, including companies like iGama, Avenga, SBRobotiks, and Intellectsoft.
Mark I Perceptron
The Mark I Perceptron was a pioneering supervised image classification learning system developed by Frank Rosenblatt in 1958. It was the first implementation of an artificial intelligence (AI) machine. It differs from the Perceptron which is a software architecture proposed in 1943 by Warren McCulloch and Walter Pitts, which was also employed in Mark I, and enhancements of which have continued to be an integral part of cutting edge AI technologies like the Transformer. == Architecture == The Mark I Perceptron was organized into three layers: A set of sensory units which receive optical input A set of association units, each of which fire based on input from multiple sensory units A set of response units, which fire based on input from multiple association units The connection between sensory units and association units were random. The working of association units was very similar to the response units. Different versions of the Mark I used different numbers of units in each of the layers. == Capabilities == In his 1957 proposal for funding for development of the "Cornell Photoperceptron", Rosenblatt claimed:"Devices of this sort are expected ultimately to be capable of concept formation, language translation, collation of military intelligence, and the solution of problems through inductive logic."With the first version of the Mark I Perceptron as early as 1958, Rosenblatt demonstrated a simple binary classification experiment, namely distinguishing between sheets of paper marked on the right versus those marked on the left side. One of the later experiments distinguished a square from a circle printed on paper. The shapes were perfect and their sizes fixed; the only variation was in their position and orientation. The Mark I Perceptron achieved 99.8% accuracy on a test dataset with 500 neurons in a single layer. The size of the training dataset was 10,000 example images. It took 3 seconds for the training pipeline to go through a single image. Higher accuracy was observed with thick outline figures compared to solid figures, likely because outline figures reduced overfitting. Another experiment distinguished between a square and a diamond for which 100% accuracy was achieved with only 60 training images, with a Perceptron having 1,000 neurons in a single layer. The time taken to process each training input for this larger perceptron was 15 seconds. The only variation was in position of the image, since rotation would have been ambiguous. In that same experiment, it could distinguish between the letters X and E with 100% accuracy when trained with only 20 images (10 images of each letter). Variations in the images included both position and rotation by up to 30 degrees. When variation in rotation was increased to any angle (both in training and test datasets), the accuracy reduced to 90% with 60 training images (30 images of each letter). For distinguishing between the letters E and F, a more challenging problem due to their similarity, the same 1,000 neuron perceptron achieved an accuracy of more than 80% with 60 training images. Variation was only in the position of the image, with no rotation.
SmartAction
SmartAction Company LLC is a U.S.-based software company that develops artificial intelligence–driven virtual agents for customer service applications, including voice-based interactive voice response (IVR) systems, chat, and SMS. The company was founded in 2009 by inventor and entrepreneur Peter Voss and is headquartered in Fort Worth, Texas. == History == In 2001, Peter Voss founded Adaptive AI, Inc., a research and development company focused on artificial intelligence concepts. In 2009, Voss founded SmartAction Company, LLC to commercialize customer-service automation software derived from this work. The company’s initial products focused on automating inbound and outbound calls for contact center environments. In November 2022, Kyle Johnson was appointed chief executive officer, succeeding Gary Davis, who had served as CEO since 2020. In 2024, SmartAction was acquired by Capacity, an AI-powered customer support automation company based in St. Louis, Missouri. == Technology == SmartAction develops cloud-based voice automation software that integrates speech recognition and natural language processing to support automated customer interactions in contact center environments. The platform supports automated handling of common customer service tasks and is designed to integrate with enterprise systems.
Oriented energy filters
Oriented energy filters are used to grant sight to intelligent machines and sensors. The light comes in and is filtered so that it can be properly computed and analyzed by the computer allowing it to “perceive” what it is measuring. These energy measurements are then calculated to take a real time measurement of the oriented space time structure. 3D Gaussian filters are used to extract orientation measurements. They were chosen due to their ability to capture a broad spectrum and easy and efficient computations. The use of these vision systems can then be used in smart room, human interface and surveillance applications. The computations used can tell more than the standalone frame that most perceived motion devices such as a television frame. The objects captured by these devices would tell the velocity and energy of an object and its direction in relation to space and time. This also allows for better tracking ability and recognition.
Convolutional neural network
A convolutional neural network (CNN) is a type of feedforward neural network that learns features via filter (or kernel) optimization. This type of deep learning network has been applied to process and make predictions from many different types of data including text, images and audio. CNNs are the de-facto standard in deep learning-based approaches to computer vision and image processing, and have only recently been replaced—in some cases—by newer architectures such as the transformer. Vanishing gradients and exploding gradients, seen during backpropagation in earlier neural networks, are prevented by the regularization that comes from using shared weights over fewer connections. For example, for each neuron in the fully-connected layer, 10,000 weights would be required for processing an image sized 100 × 100 pixels. However, applying cascaded convolution (or cross-correlation) kernels, only 25 weights for each convolutional layer are required to process 5x5-sized tiles. Higher-layer features are extracted from wider context windows, compared to lower-layer features. Some applications of CNNs include: image and video recognition, recommender systems, image classification, image segmentation, medical image analysis, natural language processing, brain–computer interfaces, and financial time series. CNNs are also known as shift invariant or space invariant artificial neural networks, based on the shared-weight architecture of the convolution kernels or filters that slide along input features and provide translation-equivariant responses known as feature maps. Counter-intuitively, most convolutional neural networks are not invariant to translation, due to the downsampling operation they apply to the input. Feedforward neural networks are usually fully connected networks, that is, each neuron in one layer is connected to all neurons in the next layer. The "full connectivity" of these networks makes them prone to overfitting data. Typical ways of regularization, or preventing overfitting, include: penalizing parameters during training (such as weight decay) or trimming connectivity (skipped connections, dropout, etc.) Robust datasets also increase the probability that CNNs will learn the generalized principles that characterize a given dataset rather than the biases of a poorly-populated set. Convolutional networks were inspired by biological processes in that the connectivity pattern between neurons resembles the organization of the animal visual cortex. Individual cortical neurons respond to stimuli only in a restricted region of the visual field known as the receptive field. The receptive fields of different neurons partially overlap such that they cover the entire visual field. CNNs use relatively little pre-processing compared to other image classification algorithms. This means that the network learns to optimize the filters (or kernels) through automated learning, whereas in traditional algorithms these filters are hand-engineered. This simplifies and automates the process, enhancing efficiency and scalability overcoming human-intervention bottlenecks. == Architecture == A convolutional neural network consists of an input layer, hidden layers and an output layer. In a convolutional neural network, the hidden layers include one or more layers that perform convolutions. Typically this includes a layer that performs a dot product of the convolution kernel with the layer's input matrix. This product is usually the Frobenius inner product, and its activation function is commonly ReLU. As the convolution kernel slides along the input matrix for the layer, the convolution operation generates a feature map, which in turn contributes to the input of the next layer. This is followed by other layers such as pooling layers, fully connected layers, and normalization layers. Here it should be noted how close a convolutional neural network is to a matched filter. === Convolutional layers === In a CNN, the input is a tensor with shape: (number of inputs) × (input height) × (input width) × (input channels) After passing through a convolutional layer, the image becomes abstracted to a feature map, also called an activation map, with shape: (number of inputs) × (feature map height) × (feature map width) × (feature map channels). Convolutional layers convolve the input and pass its result to the next layer. This is similar to the response of a neuron in the visual cortex to a specific stimulus. Each convolutional neuron processes data only for its receptive field. Although fully connected feedforward neural networks can be used to learn features and classify data, this architecture is generally impractical for larger inputs (e.g., high-resolution images), which would require massive numbers of neurons because each pixel is a relevant input feature. A fully connected layer for an image of size 100 × 100 has 10,000 weights for each neuron in the second layer. Convolution reduces the number of free parameters, allowing the network to be deeper. For example, using a 5 × 5 tiling region, each with the same shared weights, requires only 25 neurons. Using shared weights means there are many fewer parameters, which helps avoid the vanishing gradients and exploding gradients problems seen during backpropagation in earlier neural networks. To speed processing, standard convolutional layers can be replaced by depthwise separable convolutional layers, which are based on a depthwise convolution followed by a pointwise convolution. The depthwise convolution is a spatial convolution applied independently over each channel of the input tensor, while the pointwise convolution is a standard convolution restricted to the use of 1 × 1 {\displaystyle 1\times 1} kernels. === Pooling layers === Convolutional networks may include local and/or global pooling layers along with traditional convolutional layers. Pooling layers reduce the dimensions of data by combining the outputs of neuron clusters at one layer into a single neuron in the next layer. Local pooling combines small clusters, tiling sizes such as 2 × 2 are commonly used. Global pooling acts on all the neurons of the feature map. There are two common types of pooling in popular use: max and average. Max pooling uses the maximum value of each local cluster of neurons in the feature map, while average pooling takes the average value. === Fully connected layers === Fully connected layers connect every neuron in one layer to every neuron in another layer. It is the same as a traditional multilayer perceptron neural network (MLP). Each neuron in the fully connected layer receives input from all the neurons in the previous layer. These inputs are weighted and summed with the corresponding biases, and then passed through an activation function to perform a nonlinear transformation, generating the output. The flattened matrix goes through a fully connected layer to classify the images. === Receptive field === In neural networks, each neuron receives input from some number of locations in the previous layer. In a convolutional layer, each neuron receives input from only a restricted area of the previous layer called the neuron's receptive field. Typically the area is a square (e.g. 5 by 5 neurons). Whereas, in a fully connected layer, the receptive field is the entire previous layer. Thus, in each convolutional layer, each neuron takes input from a larger area in the input than previous layers. This is due to applying the convolution over and over, which takes the value of a pixel into account, as well as its surrounding pixels. When using dilated layers, the number of pixels in the receptive field remains constant, but the field is more sparsely populated as its dimensions grow when combining the effect of several layers. To manipulate the receptive field size as desired, there are some alternatives to the standard convolutional layer. For example, atrous or dilated convolution expands the receptive field size without increasing the number of parameters by interleaving visible and blind regions. Moreover, a single dilated convolutional layer can comprise filters with multiple dilation ratios, thus having a variable receptive field size. === Weights === Each neuron in a neural network computes an output value by applying a specific function to the input values received from the receptive field in the previous layer. The function that is applied to the input values is determined by a vector of weights and a bias (typically real numbers). Learning consists of iteratively adjusting these biases and weights. The vectors of weights and biases are called filters and represent particular features of the input (e.g., a particular shape). A distinguishing feature of CNNs is that many neurons can share the same filter. This reduces the memory footprint because a single bias and a single vector of weights are used across all receptive fields that share that filter, as opposed to each receptive field having its own bias and vector
European Conference on Artificial Intelligence
The European Conference on Artificial Intelligence (ECAI) is the leading conference in the field of Artificial Intelligence in Europe, and is commonly listed together with IJCAI and AAAI as one of the three major general AI conferences worldwide. The conference series has been held without interruption since 1974, originally under the name AISB. The conference was originally held biennially, but has been organized annually since ECAI 2022. The conferences are held under the auspices of the European Coordinating Committee for Artificial Intelligence (ECCAI) and organized by one of the member societies. The journal AI Communications, sponsored by the same society, regularly publishes special issues in which conference attendees report on the conference. Publication of a paper in ECAI is considered by some journals to be archival: the paper should be considered equivalent to a journal publication and that the contents of ECAI papers cannot be reformulated as separate journal submissions unless a significant amount of new material is added. == List of ECAI conferences == ECAI-1992 took place in Vienna, Austria. ECAI-1996 took place in Budapest, Hungary. ECAI-1998 tool place in Brighton, United Kingdom. ECAI-2000 took place in Berlin, Germany. ECAI-2004 took place in Valencia, Spain. ECAI-2006 took place in Riva del Garda, Italy. ECAI-2008 took place in Patras, Greece. ECAI-2010 took place in Lisbon, Portugal. ECAI-2012 took place in Montpellier, France. ECAI-2014 took place in Prague, Czech Republic. ECAI-2016 took place in The Hague, Netherlands. ECAI-2018 took place in Stockholm, Sweden. ECAI-2020 took place in Santiago de Compostela, Spain. ECAI-2022 took place in Vienna, Austria. ECAI-2023 took place in Kraków, Poland. ECAI-2024 took place in Santiago de Compostela, Spain. ECAI-2025 took place in Bologna, Italy.