Flok (formerly Loyalblocks) was an American tech startup based in New York City that provides marketing services such as chatbots/AI, customer loyalty programs, mobile apps and CRM services to local businesses. In January 2017, the company was acquired by Wix.com. Around March 2017, Flok ceased regular communication. At some point in 2019 Flok communicated to its customers that it would shut down in March 2020. == Background == Flok was founded in 2011 by Ido Gaver and Eran Kirshenboim and has offices in Tel Aviv, Israel. In May 2013, Flok secured a $9 million Series A Round from General Catalyst Partners with participation from Founder Collective and existing investor Gemini Israel Ventures. In total, Flok has raised over $18 million in venture capital in three rounds. In May 2014, Flok announced a self-service loyalty platform for SMBs to build their own programs with beacon integration. At that time, approximately 40,000 businesses were using the service. In 2016, Flok released a turnkey chatbot service for local businesses, and was featured in AdWeek for developing the first weed bot chatbot for a California cannabis business. == Services == Flok offered an eponymous customer-facing app, that consumers use to receive rewards and deals from partner businesses, and a Flok business app for merchants to manage the platform.
Data preprocessing
Data preprocessing can refer to manipulation, filtration or augmentation of data before it is analyzed, and is often an important step in the data mining process. Data collection methods are often loosely controlled, resulting in out-of-range values, impossible data combinations, and missing values, amongst other issues. Preprocessing is the process by which unstructured data is transformed into intelligible representations suitable for machine-learning models. This phase of model deals with noise in order to arrive at better and improved results from the original data set which was noisy. This dataset also has some level of missing value present in it. The preprocessing pipeline used can often have large effects on the conclusions drawn from the downstream analysis. Thus, representation and quality of data is necessary before running any analysis. If there is a high proportion of irrelevant and redundant information present or noisy and unreliable data, then knowledge discovery during the training phase may be more difficult. Data preparation and filtering steps can take a considerable amount of processing time. Examples of methods used in data preprocessing include cleaning, instance selection, normalization, one-hot encoding, data transformation, feature extraction and feature selection. == Applications == === Data mining === Data preprocessing allows for the removal of unwanted data with the use of data cleaning, this allows the user to have a dataset to contain more valuable information after the preprocessing stage for data manipulation later in the data mining process. Editing such dataset to either correct data corruption or human error is a crucial step to get accurate quantifiers like true positives, true negatives, false positives and false negatives found in a confusion matrix that are commonly used for a medical diagnosis. Users are able to join data files together and use preprocessing to filter any unnecessary noise from the data which can allow for higher accuracy. Users use Python programming scripts accompanied by the pandas library which gives them the ability to import data from a comma-separated values as a data-frame. The data-frame is then used to manipulate data that can be challenging otherwise to do in Excel. Pandas (software) which is a powerful tool that allows for data analysis and manipulation; which makes data visualizations, statistical operations and much more, a lot easier. Many also use the R programming language to do such tasks as well. The reason why a user transforms existing files into a new one is because of many reasons. Aspects of data preprocessing may include imputing missing values, aggregating numerical quantities and transforming continuous data into categories (data binning). More advanced techniques like principal component analysis and feature selection are working with statistical formulas and are applied to complex datasets which are recorded by GPS trackers and motion capture devices. === Semantic data preprocessing === Semantic data mining is a subset of data mining that specifically seeks to incorporate domain knowledge, such as formal semantics, into the data mining process. Domain knowledge is the knowledge of the environment the data was processed in. Domain knowledge can have a positive influence on many aspects of data mining, such as filtering out redundant or inconsistent data during the preprocessing phase. Domain knowledge also works as constraint. It does this by using working as set of prior knowledge to reduce the space required for searching and acting as a guide to the data. Simply put, semantic preprocessing seeks to filter data using the original environment of said data more correctly and efficiently. There are increasingly complex problems which are asking to be solved by more elaborate techniques to better analyze existing information. Instead of creating a simple script for aggregating different numerical values into a single value, it make sense to focus on semantic based data preprocessing. The idea is to build a dedicated ontology, which explains on a higher level what the problem is about. In regards to semantic data mining and semantic pre-processing, ontologies are a way to conceptualize and formally define semantic knowledge and data. The Protégé (software) is the standard tool for constructing an ontology. In general, the use of ontologies bridges the gaps between data, applications, algorithms, and results that occur from semantic mismatches. As a result, semantic data mining combined with ontology has many applications where semantic ambiguity can impact the usefulness and efficiency of data systems. Applications include the medical field, language processing, banking, and even tutoring, among many more. There are various strengths to using a semantic data mining and ontological based approach. As previously mentioned, these tools can help during the per-processing phase by filtering out non-desirable data from the data set. Additionally, well-structured formal semantics integrated into well designed ontologies can return powerful data that can be easily read and processed by machines. A specifically useful example of this exists in the medical use of semantic data processing. As an example, a patient is having a medical emergency and is being rushed to hospital. The emergency responders are trying to figure out the best medicine to administer to help the patient. Under normal data processing, scouring all the patient’s medical data to ensure they are getting the best treatment could take too long and risk the patients’ health or even life. However, using semantically processed ontologies, the first responders could save the patient’s life. Tools like a semantic reasoner can use ontology to infer the what best medicine to administer to the patient is based on their medical history, such as if they have a certain cancer or other conditions, simply by examining the natural language used in the patient's medical records. This would allow the first responders to quickly and efficiently search for medicine without having worry about the patient’s medical history themselves, as the semantic reasoner would already have analyzed this data and found solutions. In general, this illustrates the incredible strength of using semantic data mining and ontologies. They allow for quicker and more efficient data extraction on the user side, as the user has fewer variables to account for, since the semantically pre-processed data and ontology built for the data have already accounted for many of these variables. However, there are some drawbacks to this approach. Namely, it requires a high amount of computational power and complexity, even with relatively small data sets. This could result in higher costs and increased difficulties in building and maintaining semantic data processing systems. This can be mitigated somewhat if the data set is already well organized and formatted, but even then, the complexity is still higher when compared to standard data processing. Below is a simple a diagram combining some of the processes, in particular semantic data mining and their use in ontology. The diagram depicts a data set being broken up into two parts: the characteristics of its domain, or domain knowledge, and then the actual acquired data. The domain characteristics are then processed to become user understood domain knowledge that can be applied to the data. Meanwhile, the data set is processed and stored so that the domain knowledge can applied to it, so that the process may continue. This application forms the ontology. From there, the ontology can be used to analyze data and process results. Fuzzy preprocessing is another, more advanced technique for solving complex problems. Fuzzy preprocessing and fuzzy data mining make use of fuzzy sets. These data sets are composed of two elements: a set and a membership function for the set which comprises 0 and 1. Fuzzy preprocessing uses this fuzzy data set to ground numerical values with linguistic information. Raw data is then transformed into natural language. Ultimately, fuzzy data mining's goal is to help deal with inexact information, such as an incomplete database. Currently fuzzy preprocessing, as well as other fuzzy based data mining techniques see frequent use with neural networks and artificial intelligence.
Application-release automation
Application-release automation (ARA) refers to the process of packaging and deploying an application or update of an application from development, across various environments, and ultimately to production. ARA solutions must combine the capabilities of deployment automation, environment management and modeling, and release coordination. == Relationship with DevOps == ARA tools help cultivate DevOps best practices by providing a combination of automation, environment modeling and workflow-management capabilities. These practices help teams deliver software rapidly, reliably and responsibly. ARA tools achieve a key DevOps goal of implementing continuous delivery with a large quantity of releases quickly. == Relationship with deployment == ARA is more than just software-deployment automation – it deploys applications using structured release-automation techniques that allow for an increase in visibility for the whole team. It combines workload automation and release-management tools as they relate to release packages, as well as movement through different environments within the DevOps pipeline. ARA tools help regulate deployments, how environments are created and deployed, and how and when releases are deployed. == ARA Solutions == All ARA solutions must include capabilities in automation, environment modeling, and release coordination. Additionally, the solution must provide this functionality without reliance on other tools.
Automatic summarization
Automatic summarization is the process of shortening a set of data computationally, to create a subset (a summary) that represents the most important or relevant information within the original content. Artificial intelligence (AI) algorithms are commonly developed and employed to achieve this, specialized for different types of data. Text summarization is usually implemented by natural language processing methods, designed to locate the most informative sentences in a given document. On the other hand, visual content can be summarized using computer vision algorithms. Image summarization is the subject of ongoing research; existing approaches typically attempt to display the most representative images from a given image collection, or generate a video that only includes the most important content from the entire collection. Video summarization algorithms identify and extract from the original video content the most important frames (key-frames), and/or the most important video segments (key-shots), normally in a temporally ordered fashion. Video summaries simply retain a carefully selected subset of the original video frames and, therefore, are not identical to the output of video synopsis algorithms, where new video frames are being synthesized based on the original video content. == Commercial products == In 2022 Google Docs released an automatic summarization feature. == Approaches == There are two general approaches to automatic summarization: extraction and abstraction. === Extraction-based summarization === Here, content is extracted from the original data, but the extracted content is not modified in any way. Examples of extracted content include key-phrases that can be used to "tag" or index a text document, or key sentences (including headings) that collectively comprise an abstract, and representative images or video segments, as stated above. For text, extraction is analogous to the process of skimming, where the summary (if available), headings and subheadings, figures, the first and last paragraphs of a section, and optionally the first and last sentences in a paragraph are read before one chooses to read the entire document in detail. Other examples of extraction that include key sequences of text in terms of clinical relevance (including patient/problem, intervention, and outcome). === Abstractive-based summarization === Abstractive summarization methods generate new text that did not exist in the original text. This has been applied mainly for text. Abstractive methods build an internal semantic representation of the original content (often called a language model), and then use this representation to create a summary that is closer to what a human might express. Abstraction may transform the extracted content by paraphrasing sections of the source document, to condense a text more strongly than extraction. Such transformation, however, is computationally much more challenging than extraction, involving both natural language processing and often a deep understanding of the domain of the original text in cases where the original document relates to a special field of knowledge. "Paraphrasing" is even more difficult to apply to images and videos, which is why most summarization systems are extractive. === Aided summarization === Approaches aimed at higher summarization quality rely on combined software and human effort. In Machine Aided Human Summarization, extractive techniques highlight candidate passages for inclusion (to which the human adds or removes text). In Human Aided Machine Summarization, a human post-processes software output, in the same way that one edits the output of automatic translation by Google Translate. == Applications and systems for summarization == There are broadly two types of extractive summarization tasks depending on what the summarization program focuses on. The first is generic summarization, which focuses on obtaining a generic summary or abstract of the collection (whether documents, or sets of images, or videos, news stories etc.). The second is query relevant summarization, sometimes called query-based summarization, which summarizes objects specific to a query. Summarization systems are able to create both query relevant text summaries and generic machine-generated summaries depending on what the user needs. An example of a summarization problem is document summarization, which attempts to automatically produce an abstract from a given document. Sometimes one might be interested in generating a summary from a single source document, while others can use multiple source documents (for example, a cluster of articles on the same topic). This problem is called multi-document summarization. A related application is summarizing news articles. Imagine a system, which automatically pulls together news articles on a given topic (from the web), and concisely represents the latest news as a summary. Image collection summarization is another application example of automatic summarization. It consists in selecting a representative set of images from a larger set of images. A summary in this context is useful to show the most representative images of results in an image collection exploration system. Video summarization is a related domain, where the system automatically creates a trailer of a long video. This also has applications in consumer or personal videos, where one might want to skip the boring or repetitive actions. Similarly, in surveillance videos, one would want to extract important and suspicious activity, while ignoring all the boring and redundant frames captured. At a very high level, summarization algorithms try to find subsets of objects (like set of sentences, or a set of images), which cover information of the entire set. This is also called the core-set. These algorithms model notions like diversity, coverage, information and representativeness of the summary. Query based summarization techniques, additionally model for relevance of the summary with the query. Some techniques and algorithms which naturally model summarization problems are TextRank and PageRank, Submodular set function, Determinantal point process, maximal marginal relevance (MMR) etc. === Keyphrase extraction === The task is the following. You are given a piece of text, such as a journal article, and you must produce a list of keywords or key[phrase]s that capture the primary topics discussed in the text. In the case of research articles, many authors provide manually assigned keywords, but most text lacks pre-existing keyphrases. For example, news articles rarely have keyphrases attached, but it would be useful to be able to automatically do so for a number of applications discussed below. Consider the example text from a news article: "The Army Corps of Engineers, rushing to meet President Bush's promise to protect New Orleans by the start of the 2006 hurricane season, installed defective flood-control pumps last year despite warnings from its own expert that the equipment would fail during a storm, according to documents obtained by The Associated Press". A keyphrase extractor might select "Army Corps of Engineers", "President Bush", "New Orleans", and "defective flood-control pumps" as keyphrases. These are pulled directly from the text. In contrast, an abstractive keyphrase system would somehow internalize the content and generate keyphrases that do not appear in the text, but more closely resemble what a human might produce, such as "political negligence" or "inadequate protection from floods". Abstraction requires a deep understanding of the text, which makes it difficult for a computer system. Keyphrases have many applications. They can enable document browsing by providing a short summary, improve information retrieval (if documents have keyphrases assigned, a user could search by keyphrase to produce more reliable hits than a full-text search), and be employed in generating index entries for a large text corpus. Depending on the different literature and the definition of key terms, words or phrases, keyword extraction is a highly related theme. ==== Supervised learning approaches ==== Beginning with the work of Turney, many researchers have approached keyphrase extraction as a supervised machine learning problem. Given a document, we construct an example for each unigram, bigram, and trigram found in the text (though other text units are also possible, as discussed below). We then compute various features describing each example (e.g., does the phrase begin with an upper-case letter?). We assume there are known keyphrases available for a set of training documents. Using the known keyphrases, we can assign positive or negative labels to the examples. Then we learn a classifier that can discriminate between positive and negative examples as a function of the features. Some classifiers make a binary classification for a test example, while others assign a probability of being a keyphrase. For ins
Neural processing unit
A neural processing unit (NPU), also known as an AI accelerator or deep learning processor, is a class of specialized hardware accelerator or computer system designed to accelerate artificial intelligence and machine learning applications, including artificial neural networks and computer vision. == Use == Their purpose is either to efficiently execute already trained AI models (inference) or to train AI models. NPUs can be more efficient in terms of speed or power consumption. NPU applications include algorithms for robotics, Internet of things, and data-intensive or sensor-driven tasks. They are often manycore or spatial designs and focus on low-precision arithmetic, novel dataflow architectures, or in-memory computing capability. As of 2024, a widely used datacenter-grade AI integrated circuit chip, the Nvidia H100 GPU, contains tens of billions of MOSFETs. === Consumer devices === AI accelerators are used in Apple silicon, Qualcomm, Samsung, Huawei, and Google Tensor smartphone processors. Vision processing units are accelerators specialized for machine vision algorithms such as CNN (convolutional neural networks) and SIFT (scale-invariant feature transform). They are used in devices that need to keep track of objects visually such as AR headsets and drones. It is more recently (circa 2017) added to processors from Apple and (circa 2022) to processors from Intel and AMD. All models of Intel Meteor Lake processors have a built-in versatile processor unit (VPU) for accelerating inference for computer vision and deep learning. On consumer devices, the NPU is intended to be small, power-efficient, but reasonably fast when used to run small models. To do this they are designed to support low-bitwidth operations using data types such as INT4, INT8, FP8, and FP16. A common metric is trillions of operations per second (TOPS). Although TOPS does not explicitly specify the kind of operations, it is typically INT8 additions and multiplications. === Datacenters === Accelerators are used in cloud computing servers: e.g., tensor processing units (TPU) for Google Cloud Platform, and Trainium and Inferentia chips for Amazon Web Services. Many vendor-specific terms exist for devices in this category, and it is an emerging technology without a dominant design. Since the late 2010s, graphics processing units designed by companies such as Nvidia and AMD often include AI-specific hardware in the form of dedicated functional units for low-precision matrix-multiplication operations. These GPUs are commonly used as AI accelerators, both for training and inference. === Scientific computation === Although NPUs are tailored for low-precision (e.g., FP16, INT8) matrix multiplication operations, they can be used to emulate higher-precision matrix multiplications in scientific computing. As modern GPUs place much focus on making the NPU part fast, using emulated FP64 (Ozaki scheme) on NPUs can potentially outperform native FP64. This has been demonstrated using FP16-emulated FP64 on NVIDIA TITAN RTX and using INT8-emulated FP64 on NVIDIA consumer GPUs and the A100 GPU. Consumer GPUs especially benefited as they have limited FP64 hardware capacity, showing a 6× speedup. Since CUDA Toolkit 13.0 Update 2, cuBLAS automatically uses INT8-emulated FP64 matrix multiplication of the equivalent precision if it is faster than native. This is in addition to the FP16-emulated FP32 feature introduced in version 12.9. == Programming == An operating system or a higher-level library may provide application programming interfaces such as TensorFlow with LiteRT Next (Android), CoreML (iOS, macOS) or DirectML (Windows). Formats such as ONNX are used to represent trained neural networks. Consumer CPU-integrated NPUs are accessible through vendor-specific APIs. AMD (Ryzen AI), Intel (OpenVINO), Apple silicon (CoreML), and Qualcomm (SNPE) each have their own APIs, which can be built upon by a higher-level library. GPUs generally use existing GPGPU pipelines such as CUDA and OpenCL adapted for lower precisions and specialized matrix-multiplication operations. Vulkan is also being used. Custom-built systems such as the Google TPU use private interfaces. There are a large number of separate underlying acceleration APIs and compilers/runtimes in use in the AI field, causing a great increase in software development effort due to the many combinations involved. As of 2025, the open standard organization Khronos Group is pursuing standardization of AI-related interfaces to reduce the amount of work needed. Khronos is working on three separate fronts: expansion of data types and intrinsic operations in OpenCL and Vulkan, inclusion of compute graphs in SPIR-V, and a NNEF/SkriptND file format for describing a neural network.
Observability (software)
In software engineering, more specifically in distributed computing, observability is the ability to collect data about programs' execution, modules' internal states, and the communication among components. To improve observability, software engineers use a wide range of logging and tracing techniques to gather telemetry information, and tools to analyze and use it. Observability is foundational to site reliability engineering, as it is the first step in triaging a service outage. One of the goals of observability is to minimize the amount of prior knowledge needed to debug an issue. == Etymology, terminology and definition == The term is borrowed from control theory, where the "observability" of a system measures how well its state can be determined from its outputs. Similarly, software observability measures how well a system's state can be understood from the obtained telemetry (metrics, logs, traces, profiling). The definition of observability varies by vendor: Observability is the process of making a system’s internal state more transparent. Systems are made observable by the data they produce, which in turn helps you to determine if your infrastructure or application is healthy and functioning normally. a measure of how well you can understand and explain any state your system can get into, no matter how novel or bizarre [...] without needing to ship new code software tools and practices for aggregating, correlating and analyzing a steady stream of performance data from a distributed application along with the hardware and network it runs onobservability starts by shipping all your raw data to central service before you begin analysisthe ability to measure a system’s current state based on the data it generates, such as logs, metrics, and traces Observability is tooling or a technical solution that allows teams to actively debug their system. Observability is based on exploring properties and patterns not defined in advance. proactively collecting, visualizing, and applying intelligence to all of your metrics, events, logs, and traces—so you can understand the behavior of your complex digital system The term is frequently referred to as its numeronym o11y (where 11 stands for the number of letters between the first letter and the last letter of the word). This is similar to other computer science abbreviations such as i18n and l10n and k8s. === Observability vs. monitoring === Observability and monitoring are sometimes used interchangeably. As tooling, commercial offerings and practices evolved in complexity, "monitoring" was re-branded as observability in order to differentiate new tools from the old. The terms are commonly contrasted in that systems are monitored using predefined sets of telemetry, and monitored systems may be observable. Majors et al. suggest that engineering teams that only have monitoring tools end up relying on expert foreknowledge (seniority), whereas teams that have observability tools rely on exploratory analysis (curiosity). == Telemetry types == Observability relies on three main types of telemetry data: metrics, logs and traces. Those are often referred to as "pillars of observability". === Metrics === A metric is a point in time measurement (scalar) that represents some system state. Examples of common metrics include: number of HTTP requests per second; total number of query failures; database size in bytes; time in seconds since last garbage collection. Monitoring tools are typically configured to emit alerts when certain metric values exceed set thresholds. Thresholds are set based on knowledge about normal operating conditions and experience. Metrics are typically tagged to facilitate grouping and searchability. Application developers choose what kind of metrics to instrument their software with, before it is released. As a result, when a previously unknown issue is encountered, it is impossible to add new metrics without shipping new code. Furthermore, their cardinality can quickly make the storage size of telemetry data prohibitively expensive. Since metrics are cardinality-limited, they are often used to represent aggregate values (for example: average page load time, or 5-second average of the request rate). Without external context, it is impossible to correlate between events (such as user requests) and distinct metric values. === Logs === Logs, or log lines, are generally free-form, unstructured text blobs that are intended to be human readable. Modern logging is structured to enable machine parsability. As with metrics, an application developer must instrument the application upfront and ship new code if different logging information is required. Logs typically include a timestamp and severity level. An event (such as a user request) may be fragmented across multiple log lines and interweave with logs from concurrent events. === Traces === ==== Distributed traces ==== A cloud native application is typically made up of distributed services which together fulfill a single request. A distributed trace is an interrelated series of discrete events (also called spans) that track the progression of a single user request. A trace shows the causal and temporal relationships between the services that interoperate to fulfill a request. Instrumenting an application with traces means sending span information to a tracing backend. The tracing backend correlates the received spans to generate presentable traces. To be able to follow a request as it traverses multiple services, spans are labeled with unique identifiers that enable constructing a parent-child relationship between spans. Span information is typically shared in the HTTP headers of outbound requests. === Continuous profiling === Continuous profiling is another telemetry type used to precisely determine how an application consumes resources. === Instrumentation === To be able to observe an application, telemetry about the application's behavior needs to be collected or exported. Instrumentation means generating telemetry alongside the normal operation of the application. Telemetry is then collected by an independent backend for later analysis. In fast-changing systems, instrumentation itself is often the best possible documentation, since it combines intention (what are the dimensions that an engineer named and decided to collect?) with the real-time, up-to-date information of live status in production. Instrumentation can be automatic, or custom. Automatic instrumentation offers blanket coverage and immediate value; custom instrumentation brings higher value but requires more intimate involvement with the instrumented application. Instrumentation can be native - done in-code (modifying the code of the instrumented application) - or out-of-code (e.g. sidecar, eBPF). Verifying new features in production by shipping them together with custom instrumentation is a practice called "observability-driven development". == "Pillars of observability" == Metrics, logs and traces are most commonly listed as the pillars of observability. Majors et al. suggest that the pillars of observability are high cardinality, high-dimensionality, and explorability, arguing that runbooks and dashboards have little value because "modern systems rarely fail in precisely the same way twice." == Self monitoring == Self monitoring is a practice where observability stacks monitor each other, in order to reduce the risk of inconspicuous outages. Self monitoring may be put in place in addition to high availability and redundancy to further avoid correlated failures.
Deep learning
In machine learning, deep learning (DL) focuses on utilizing multilayered neural networks to perform tasks such as classification, regression, and representation learning. The field takes inspiration from biological neuroscience and revolves around stacking artificial neurons into layers and "training" them to process data. The adjective "deep" refers to the use of multiple layers (ranging from three to several hundred or thousands) in the network. Methods used can be supervised, semi-supervised or unsupervised. Some common deep learning network architectures include fully connected networks, deep belief networks, recurrent neural networks, convolutional neural networks, generative adversarial networks, transformers, and neural radiance fields. These architectures have been applied to fields including computer vision, speech recognition, natural language processing, machine translation, bioinformatics, drug design, medical image analysis, climate science, material inspection and board game programs, where they have produced results comparable to and in some cases surpassing human expert performance. Early forms of neural networks were inspired by information processing and distributed communication nodes in biological systems, particularly the human brain. However, current neural networks do not intend to model the brain function of organisms, and are generally seen as low-quality models for that purpose. == Overview == Most modern deep learning models are based on multi-layered neural networks such as convolutional neural networks and transformers, although they can also include propositional formulas or latent variables organized layer-wise in deep generative models such as the nodes in deep belief networks and deep Boltzmann machines. Fundamentally, deep learning refers to a class of machine learning algorithms in which a hierarchy of layers is used to transform input data into a progressively more abstract and composite representation. For example, in an image recognition model, the raw input may be an image (represented as a tensor of pixels). The first representational layer may attempt to identify basic shapes such as lines and circles, the second layer may compose and encode arrangements of edges, the third layer may encode a nose and eyes, and the fourth layer may recognize that the image contains a face. Importantly, a deep learning process can learn which features to optimally place at which level on its own. Prior to deep learning, machine learning techniques often involved hand-crafted feature engineering to transform the data into a more suitable representation for a classification algorithm to operate on. In the deep learning approach, features are not hand-crafted and the model discovers useful feature representations from the data automatically. This does not eliminate the need for hand-tuning; for example, varying numbers of layers and layer sizes can provide different degrees of abstraction. The word "deep" in "deep learning" refers to the number of layers through which the data is transformed. More precisely, deep learning systems have a substantial credit assignment path (CAP) depth. The CAP is the chain of transformations from input to output. CAPs describe potentially causal connections between input and output. For a feedforward neural network, the depth of the CAPs is that of the network and is the number of hidden layers plus one (as the output layer is also parameterized). For recurrent neural networks, in which a signal may propagate through a layer more than once, the CAP depth is potentially unlimited. No universally agreed-upon threshold of depth divides shallow learning from deep learning, but most researchers agree that deep learning involves CAP depth higher than two. CAP of depth two has been shown to be a universal approximator in the sense that it can emulate any function. Beyond that, more layers do not add to the function approximator ability of the network. Deep models (CAP > two) are able to extract better features than shallow models and hence, extra layers help in learning the features effectively. Deep learning architectures can be constructed with a greedy layer-by-layer method. Deep learning helps to disentangle these abstractions and pick out which features improve performance. Deep learning algorithms can be applied to unsupervised learning tasks. This is an important benefit because unlabeled data is more abundant than labeled data. Examples of deep structures that can be trained in an unsupervised manner are deep belief networks. The term deep learning was introduced to the machine learning community by Rina Dechter in 1986, and to artificial neural networks by Igor Aizenberg and colleagues in 2000, in the context of Boolean threshold neurons. The etymology of the term is more complicated. == Interpretations == Deep neural networks are generally interpreted in terms of the universal approximation theorem or probabilistic inference. The classic universal approximation theorem concerns the capacity of feedforward neural networks with a single hidden layer of finite size to approximate continuous functions. In 1989, the first proof was published by George Cybenko for sigmoid activation functions and was generalised to feed-forward multi-layer architectures in 1991 by Kurt Hornik. Recent work also showed that universal approximation also holds for non-bounded activation functions such as Kunihiko Fukushima's rectified linear unit. The universal approximation theorem for deep neural networks concerns the capacity of networks with bounded width but the depth is allowed to grow. Lu et al. proved that if the width of a deep neural network with ReLU activation is strictly larger than the input dimension, then the network can approximate any Lebesgue integrable function; if the width is smaller or equal to the input dimension, then a deep neural network is not a universal approximator. The probabilistic interpretation derives from the field of machine learning. It features inference, as well as the optimization concepts of training and testing, related to fitting and generalization, respectively. More specifically, the probabilistic interpretation considers the activation nonlinearity as a cumulative distribution function. The probabilistic interpretation led to the introduction of dropout as regularizer in neural networks. The probabilistic interpretation was introduced by researchers including Hopfield, Widrow and Narendra and popularized in surveys such as the one by Bishop. == History == === Before 1980 === There are two types of artificial neural network (ANN): feedforward neural network (FNN) or multilayer perceptron (MLP) and recurrent neural networks (RNN). RNNs have cycles in their connectivity structure, whereas FNNs do not. In the 1920s, Wilhelm Lenz and Ernst Ising created the Ising model which is essentially a non-learning RNN architecture consisting of neuron-like threshold elements. In 1972, Shun'ichi Amari made this architecture adaptive. His learning RNN was republished by John Hopfield in 1982. Other early recurrent neural networks were published by Kaoru Nakano in 1971. Already in 1948, Alan Turing produced work on "Intelligent Machinery" that was not published in his lifetime, containing "ideas related to artificial evolution and learning RNNs". Frank Rosenblatt (1958) proposed the perceptron, an MLP with 3 layers: an input layer, a hidden layer with randomized weights that did not learn, and an output layer. He later published a 1962 book that also introduced variants and computer experiments, including a version with four-layer perceptrons "with adaptive preterminal networks" where the last two layers have learned weights (here he credits H. D. Block and B. W. Knight). The book cites an earlier network by R. D. Joseph (1960) "functionally equivalent to a variation of" this four-layer system (the book mentions Joseph over 30 times). Should Joseph therefore be considered the originator of proper adaptive multilayer perceptrons with learning hidden units? Unfortunately, the learning algorithm was not a functional one, and fell into oblivion. The first working deep learning algorithm was the Group method of data handling, a method to train arbitrarily deep neural networks, published by Alexey Ivakhnenko and Lapa in 1965. They regarded it as a form of polynomial regression, or a generalization of Rosenblatt's perceptron to handle more complex, nonlinear, and hierarchical relationships. A 1971 paper described a deep network with eight layers trained by this method, which is based on layer by layer training through regression analysis. Superfluous hidden units are pruned using a separate validation set. Since the activation functions of the nodes are Kolmogorov-Gabor polynomials, these were also the first deep networks with multiplicative units or "gates". The first deep learning multilayer perceptron trained by stochastic gradient descent was published in 1967 by Shun'ichi