Sparse dictionary learning (also known as sparse coding or SDL) is a representation learning method which aims to find a sparse representation of the input data in the form of a linear combination of basic elements as well as those basic elements themselves. These elements are called atoms, and they compose a dictionary. Atoms in the dictionary are not required to be orthogonal, and they may be an over-complete spanning set. This problem setup also allows the dimensionality of the signals being represented to be higher than any one of the signals being observed. These two properties lead to having seemingly redundant atoms that allow multiple representations of the same signal, but also provide an improvement in sparsity and flexibility of the representation. One of the most important applications of sparse dictionary learning is in the field of compressed sensing or signal recovery. In compressed sensing, a high-dimensional signal can be recovered with only a few linear measurements, provided that the signal is sparse or near-sparse. Since not all signals satisfy this condition, it is crucial to find a sparse representation of that signal such as the wavelet transform or the directional gradient of a rasterized matrix. Once a matrix or a high-dimensional vector is transferred to a sparse space, different recovery algorithms like basis pursuit, CoSaMP, or fast non-iterative algorithms can be used to recover the signal. One of the key principles of dictionary learning is that the dictionary has to be inferred from the input data. The emergence of sparse dictionary learning methods was stimulated by the fact that in signal processing, one typically wants to represent the input data using a minimal amount of components. Before this approach, the general practice was to use predefined dictionaries such as Fourier or wavelet transforms. However, in certain cases, a dictionary that is trained to fit the input data can significantly improve the sparsity, which has applications in data decomposition, compression, and analysis, and has been used in the fields of image denoising and classification, and video and audio processing. Sparsity and overcomplete dictionaries have immense applications in image compression, image fusion, and inpainting. == Problem statement == Given the input dataset X = [ x 1 , . . . , x K ] , x i ∈ R d {\displaystyle X=[x_{1},...,x_{K}],x_{i}\in \mathbb {R} ^{d}} we wish to find a dictionary D ∈ R d × n : D = [ d 1 , . . . , d n ] {\displaystyle \mathbf {D} \in \mathbb {R} ^{d\times n}:D=[d_{1},...,d_{n}]} and a representation R = [ r 1 , . . . , r K ] , r i ∈ R n {\displaystyle R=[r_{1},...,r_{K}],r_{i}\in \mathbb {R} ^{n}} such that both ‖ X − D R ‖ F 2 {\displaystyle \|X-\mathbf {D} R\|_{F}^{2}} is minimized and the representations r i {\displaystyle r_{i}} are sparse enough. This can be formulated as the following optimization problem: argmin D ∈ C , r i ∈ R n ∑ i = 1 K ‖ x i − D r i ‖ 2 2 + λ ‖ r i ‖ 0 {\displaystyle {\underset {\mathbf {D} \in {\mathcal {C}},r_{i}\in \mathbb {R} ^{n}}{\text{argmin}}}\sum _{i=1}^{K}\|x_{i}-\mathbf {D} r_{i}\|_{2}^{2}+\lambda \|r_{i}\|_{0}} , where C ≡ { D ∈ R d × n : ‖ d i ‖ 2 ≤ 1 ∀ i = 1 , . . . , n } {\displaystyle {\mathcal {C}}\equiv \{\mathbf {D} \in \mathbb {R} ^{d\times n}:\|d_{i}\|_{2}\leq 1\,\,\forall i=1,...,n\}} , λ > 0 {\displaystyle \lambda >0} C {\displaystyle {\mathcal {C}}} is required to constrain D {\displaystyle \mathbf {D} } so that its atoms would not reach arbitrarily high values allowing for arbitrarily low (but non-zero) values of r i {\displaystyle r_{i}} . λ {\displaystyle \lambda } controls the trade off between the sparsity and the minimization error. The minimization problem above is not convex because of the ℓ0-"norm" and solving this problem is NP-hard. In some cases L1-norm is known to ensure sparsity and so the above becomes a convex optimization problem with respect to each of the variables D {\displaystyle \mathbf {D} } and R {\displaystyle \mathbf {R} } when the other one is fixed, but it is not jointly convex in ( D , R ) {\displaystyle (\mathbf {D} ,\mathbf {R} )} . === Properties of the dictionary === The dictionary D {\displaystyle \mathbf {D} } defined above can be "undercomplete" if n < d {\displaystyle n
Text Retrieval Conference
The Text REtrieval Conference (TREC) is an ongoing series of workshops focusing on a list of different information retrieval (IR) research areas, or tracks. It is co-sponsored by the National Institute of Standards and Technology (NIST) and the Intelligence Advanced Research Projects Activity (part of the office of the Director of National Intelligence), and began in 1992 as part of the TIPSTER Text program. Its purpose is to support and encourage research within the information retrieval community by providing the infrastructure necessary for large-scale evaluation of text retrieval methodologies and to increase the speed of lab-to-product transfer of technology. TREC's evaluation protocols have improved many search technologies. A 2010 study estimated that "without TREC, U.S. Internet users would have spent up to 3.15 billion additional hours using web search engines between 1999 and 2009." Hal Varian the Chief Economist at Google wrote that "The TREC data revitalized research on information retrieval. Having a standard, widely available, and carefully constructed set of data laid the groundwork for further innovation in this field." Each track has a challenge wherein NIST provides participating groups with data sets and test problems. Depending on track, test problems might be questions, topics, or target extractable features. Uniform scoring is performed so the systems can be fairly evaluated. After evaluation of the results, a workshop provides a place for participants to collect together thoughts and ideas and present current and future research work.Text Retrieval Conference started in 1992, funded by DARPA (US Defense Advanced Research Project) and run by NIST. Its purpose was to support research within the information retrieval community by providing the infrastructure necessary for large-scale evaluation of text retrieval methodologies. == Goals == Encourage retrieval search based on large text collections Increase communication among industry, academia, and government by creating an open forum for the exchange of research ideas Speed the transfer of technology from research labs into commercial products by demonstrating substantial improvements retrieval methodologies on real world problems To increase the availability of appropriate evaluation techniques for use by industry and academia including development of new evaluation techniques more applicable to current systems TREC is overseen by a program committee consisting of representatives from government, industry, and academia. For each TREC, NIST provide a set of documents and questions. Participants run their own retrieval system on the data and return to NIST a list of retrieved top-ranked documents. NIST pools the individual result judges the retrieved documents for correctness and evaluates the results. The TREC cycle ends with a workshop that is a forum for participants to share their experiences. == Relevance judgments in TREC == TREC defines relevance as: "If you were writing a report on the subject of the topic and would use the information contained in the document in the report, then the document is relevant." Most TREC retrieval tasks use binary relevance: a document is either relevant or not relevant. Some TREC tasks use graded relevance, capturing multiple degrees of relevance. Most TREC collections are too large to perform complete relevance assessment; for these collections it is impossible to calculate the absolute recall for each query. To decide which documents to assess, TREC usually uses a method call pooling. In this method, the top-ranked n documents from each contributing run are aggregated, and the resulting document set is judged completely. == Various TRECs == In 1992 TREC-1 was held at NIST. The first conference attracted 28 groups of researchers from academia and industry. It demonstrated a wide range of different approaches to the retrieval of text from large document collections .Finally TREC1 revealed the facts that automatic construction of queries from natural language query statements seems to work. Techniques based on natural language processing were no better no worse than those based on vector or probabilistic approach. TREC2 Took place in August 1993. 31 group of researchers participated in this. Two types of retrieval were examined. Retrieval using an ‘ad hoc’ query and retrieval using a ‘routing' query In TREC-3 a small group experiments worked with Spanish language collection and others dealt with interactive query formulation in multiple databases TREC-4 they made even shorter to investigate the problems with very short user statements TREC-5 includes both short and long versions of the topics with the goal of carrying out deeper investigation into which types of techniques work well on various lengths of topics In TREC-6 Three new tracks speech, cross language, high precision information retrieval were introduced. The goal of cross language information retrieval is to facilitate research on system that are able to retrieve relevant document regardless of language of the source document TREC-7 contained seven tracks out of which two were new Query track and very large corpus track. The goal of the query track was to create a large query collection TREC-8 contain seven tracks out of which two –question answering and web tracks were new. The objective of QA query is to explore the possibilities of providing answers to specific natural language queries TREC-9 Includes seven tracks In TREC-10 Video tracks introduced Video tracks design to promote research in content based retrieval from digital video In TREC-11 Novelty tracks introduced. The goal of novelty track is to investigate systems abilities to locate relevant and new information within the ranked set of documents returned by a traditional document retrieval system TREC-12 held in 2003 added three new tracks; Genome track, robust retrieval track, HARD (Highly Accurate Retrieval from Documents) == Tracks == === Current tracks === New tracks are added as new research needs are identified, this list is current for TREC 2018. CENTRE Track – Goal: run in parallel CLEF 2018, NTCIR-14, TREC 2018 to develop and tune an IR reproducibility evaluation protocol (new track for 2018). Common Core Track – Goal: an ad hoc search task over news documents. Complex Answer Retrieval (CAR) – Goal: to develop systems capable of answering complex information needs by collating information from an entire corpus. Incident Streams Track – Goal: to research technologies to automatically process social media streams during emergency situations (new track for TREC 2018). The News Track – Goal: partnership with The Washington Post to develop test collections in news environment (new for 2018). Precision Medicine Track – Goal: a specialization of the Clinical Decision Support track to focus on linking oncology patient data to clinical trials. Real-Time Summarization Track (RTS) – Goal: to explore techniques for real-time update summaries from social media streams. === Past tracks === Chemical Track – Goal: to develop and evaluate technology for large scale search in chemistry-related documents, including academic papers and patents, to better meet the needs of professional searchers, and specifically patent searchers and chemists. Clinical Decision Support Track – Goal: to investigate techniques for linking medical cases to information relevant for patient care Contextual Suggestion Track – Goal: to investigate search techniques for complex information needs that are highly dependent on context and user interests. Crowdsourcing Track – Goal: to provide a collaborative venue for exploring crowdsourcing methods both for evaluating search and for performing search tasks. Genomics Track – Goal: to study the retrieval of genomic data, not just gene sequences but also supporting documentation such as research papers, lab reports, etc. Last ran on TREC 2007. Dynamic Domain Track – Goal: to investigate domain-specific search algorithms that adapt to the dynamic information needs of professional users as they explore in complex domains. Enterprise Track – Goal: to study search over the data of an organization to complete some task. Last ran on TREC 2008. Entity Track – Goal: to perform entity-related search on Web data. These search tasks (such as finding entities and properties of entities) address common information needs that are not that well modeled as ad hoc document search. Cross-Language Track – Goal: to investigate the ability of retrieval systems to find documents topically regardless of source language. After 1999, this track spun off into CLEF. FedWeb Track – Goal: to select best resources to forward a query to, and merge the results so that most relevant are on the top. Federated Web Search Track – Goal: to investigate techniques for the selection and combination of search results from a large number of real on-line web search services. Filtering Track – Goal: to binarily decide retrieval of new
Interplanetary Internet
The interplanetary Internet is a conceived computer network in space, consisting of a set of network nodes that can communicate with each other. These nodes are the planet's orbiters and landers, and the Earth ground stations. For example, the orbiters collect the scientific data from the Curiosity rover on Mars through near-Mars communication links, transmit the data to Earth through direct links from the Mars orbiters to the Earth ground stations via the NASA Deep Space Network, and finally the data routed through Earth's internal internet. Interplanetary communication is greatly delayed by interplanetary distances, as data transmission can only go as fast as the speed of light, so a new set of protocols and technologies that are tolerant to large delays and errors are required. The interplanetary Internet has been envisioned as a store and forward network of internets that is often disconnected, has a wireless backbone fraught with error-prone links and delays ranging from tens of minutes to even hours, even when there is a connection. As of 2024 agencies and companies working towards bringing the network to fruition include NASA, ESA, SpaceX and Blue Origin. == Challenges and reasons == In the core implementation of Interplanetary Internet, satellites orbiting a planet communicate to other planet's satellites. Simultaneously, these planets revolve around the Sun with long distances, and thus many challenges face the communications. The reasons and the resultant challenges are: The motion and long distances between planets: The interplanetary communication is greatly delayed due to the interplanetary distances and the motion of the planets. The delay is variable and long, ranging from a couple of minutes (Earth-to-Mars), to a couple of hours (Pluto-to-Earth), depending on their relative positions. The interplanetary communication also suspends due to the solar conjunction, when the sun's radiation hinders the direct communication between the planets. As such, the communication characterizes lossy links and intermittent link connectivity. Low embeddable payload: Satellites can only carry a small payload, which poses challenges to the power, mass, size, and cost for communication hardware design. An asymmetric bandwidth would be the result of this limitation. This asymmetry reaches ratios up to 1000:1 as downlink:uplink bandwidth portion. Absence of fixed infrastructure: The graph of participating nodes in a specific planet-to-planet communication keeps changing over time, due to the constant motion. The routes of the planet-to-planet communication are planned and scheduled rather than being opportunistic. The Interplanetary Internet design must address these challenges to operate successfully and achieve good communication with other planets. It also must use the few available resources efficiently in the system. == Development == Space communication technology has steadily evolved from expensive, one-of-a-kind point-to-point architectures, to the re-use of technology on successive missions, to the development of standard protocols agreed upon by space agencies of many countries. This last phase has gone on since 1982 through the efforts of the Consultative Committee for Space Data Systems (CCSDS), a body composed of the major space agencies of the world. It has 11 member agencies, 32 observer agencies, and over 119 industrial associates. The evolution of space data system standards has gone on in parallel with the evolution of the Internet, with conceptual cross-pollination where fruitful, but largely as a separate evolution. Since the late 1990s, familiar Internet protocols and CCSDS space link protocols have integrated and converged in several ways; for example, the successful FTP file transfer to Earth-orbiting STRV 1B on January 2, 1996, which ran FTP over the CCSDS IPv4-like Space Communications Protocol Specifications (SCPS) protocols. Internet Protocol use without CCSDS has taken place on spacecraft, e.g., demonstrations on the UoSAT-12 satellite, and operationally on the Disaster Monitoring Constellation. Having reached the era where networking and IP on board spacecraft have been shown to be feasible and reliable, a forward-looking study of the bigger picture was the next phase. The Interplanetary Internet study at NASA's Jet Propulsion Laboratory (JPL) was started by a team of scientists at JPL led by internet pioneer Vinton Cerf and the late Adrian Hooke. Cerf was appointed as a distinguished visiting scientist at JPL in 1998, while Hooke was one of the founders and directors of CCSDS. While IP-like SCPS protocols are feasible for short hops, such as ground station to orbiter, rover to lander, lander to orbiter, probe to flyby, and so on, delay-tolerant networking is needed to get information from one region of the Solar System to another. It becomes apparent that the concept of a region is a natural architectural factoring of the Interplanetary Internet. A region is an area where the characteristics of communication are the same. Region characteristics include communications, security, the maintenance of resources, perhaps ownership, and other factors. The Interplanetary Internet is a "network of regional internets". What is needed then, is a standard way to achieve end-to-end communication through multiple regions in a disconnected, variable-delay environment using a generalized suite of protocols. Examples of regions might include the terrestrial Internet as a region, a region on the surface of the Moon or Mars, or a ground-to-orbit region. The recognition of this requirement led to the concept of a "bundle" as a high-level way to address the generalized Store-and-Forward problem. Bundles are an area of new protocol development in the upper layers of the OSI model, above the Transport Layer with the goal of addressing the issue of bundling store-and-forward information so that it can reliably traverse radically dissimilar environments constituting a "network of regional internets". Delay-tolerant networking (DTN) was designed to enable standardized communications over long distances and through time delays. At its core is the Bundle Protocol (BP), which is similar to the Internet Protocol, or IP, that serves as the heart of the Internet here on Earth. The big difference between the regular Internet Protocol (IP) and the Bundle Protocol is that IP assumes a seamless end-to-end data path, while BP is built to account for errors and disconnections — glitches that commonly plague deep-space communications. Bundle Service Layering, implemented as the Bundling protocol suite for delay-tolerant networking, will provide general-purpose delay-tolerant protocol services in support of a range of applications: custody transfer, segmentation and reassembly, end-to-end reliability, end-to-end security, and end-to-end routing among them. The Bundle Protocol was first tested in space on the UK-DMC satellite in 2008. An example of one of these end-to-end applications flown on a space mission is the CCSDS File Delivery Protocol (CFDP), used on the Deep Impact comet mission. CFDP is an international standard for automatic, reliable file transfer in both directions. CFDP should not be confused with Coherent File Distribution Protocol, which has the same acronym and is an IETF-documented experimental protocol for rapidly deploying files to multiple targets in a highly networked environment. In addition to reliably copying a file from one entity (such as a spacecraft or ground station) to another entity, CFDP has the capability to reliably transmit arbitrarily small messages defined by the user, in the metadata accompanying the file, and to reliably transmit commands relating to file system management that are to be executed automatically on the remote end-point entity (such as a spacecraft) upon successful reception of a file. == Protocol == The Consultative Committee for Space Data Systems (CCSDS) packet telemetry standard defines the protocol used for the transmission of spacecraft instrument data over the deep-space channel. Under this standard, an image or other data sent from a spacecraft instrument is transmitted using one or more packets. === CCSDS packet definition === A packet is a block of data with length that can vary between successive packets, ranging from 7 to 65,542 bytes, including the packet header. Packetized data is transmitted via frames, which are fixed-length data blocks. The size of a frame, including frame header and control information, can range up to 2048 bytes. Packet sizes are fixed during the development phase. Because packet lengths are variable but frame lengths are fixed, packet boundaries usually do not coincide with frame boundaries. === Telecom processing notes === Data in a frame is typically protected from channel errors by error-correcting codes. Even when the channel errors exceed the correction capability of the error-correcting code, the presence of errors is nearly always detected by the e
Ciphertext expansion
In cryptography, the term ciphertext expansion refers to the length increase of a message when it is encrypted. Many modern cryptosystems cause some degree of expansion during the encryption process, for instance when the resulting ciphertext must include a message-unique Initialization Vector (IV). Probabilistic encryption schemes cause ciphertext expansion, as the set of possible ciphertexts is necessarily greater than the set of input plaintexts. Certain schemes, such as Cocks Identity Based Encryption, or the Goldwasser-Micali cryptosystem result in ciphertexts hundreds or thousands of times longer than the plaintext. Ciphertext expansion may be offset or increased by other processes which compress or expand the message, e.g., data compression or error correction coding. == Reasons why Ciphertext expansion can occur == === Probabilistic Encryption === Probabilistic encryption schemes, such as the Goldwasser-Micali cryptosystem, necessarily produce ciphertexts that are longer than the original plaintexts. This is because the set of possible ciphertexts must be larger than the set of plaintexts to achieve semantic security. === Initialization Vectors (IVs) === Many block cipher modes of operation, like Cipher Block Chaining (CBC), require the use of an Initialization Vector (IV) that is unique for each message. The IV is typically appended to the ciphertext, resulting in expansion. === Redundancy and Error Correction === Some cryptographic schemes intentionally introduce redundancy or error correction codes into the ciphertext to protect against tampering or transmission errors. This added data increases the ciphertext size. === Specific Cryptosystems === Certain cryptographic schemes, such as Cocks Identity-Based Encryption, can produce ciphertexts that are hundreds or thousands of times longer than the original plaintext. This extreme expansion is a design choice to achieve the desired security properties. Ciphertext expansion can be offset or increased by other processes that compress or expand the message, such as data compression or error correction coding. The overall impact on message size depends on the relative strengths of these competing effects.
What I eat in a day video
"What I eat in a day" videos are a trend on several social media platforms where a person describes all the meals and snacks that they eat during a given day, often as part of a given diet. The videos, shared on platforms including Twitter, TikTok and YouTube, become increasingly popular in 2020, with some of them accumulating millions of views, and they are considered a profitable industry for the people making them. Some have raised concerns that the videos may promote an unrealistic standard for healthy eating and contribute to the development of eating disorders. == Format == These videos often feature a montage of the food that the creator eats over the course of the day, sometimes with the associated calorie count of the foods that they describe. Unlike related mukbang videos, however, in which participants eat large amounts of food, the diets described are often restrictive. However, other videos are labeled as "unhealthy" and depict large portion sizes and higher amounts of processed food. == Popularity == "What I eat in a day" videos have existed for a long time, especially on YouTube, but they have become much more widespread in recent years. This phenomenon is self-reinforcing because when social media users watch or like these videos they are likely to see more of them in the future. Indeed, some of the most successful videos have tens of millions of view each. == Criticism and controversy == Several dieticians and mental health professionals over the impacts that these videos can have, as they can advocate a restrictive style of eating and not "promote body diversity." They have also raised concerns that this trend could contribute to a rise in disordered eating, especially since use of social media is known to increase feelings of negative body image. This trend is particularly prevalent among young adults, which are also the group with the highest vulnerability to eating disorders. More recently, a portion of these videos have begun to challenge diets and depict more realistic ways of eating in order to reduce the potential consequences of the trend.
Energy-based model
An energy-based model (EBM), also called Canonical Ensemble Learning (CEL) or Learning via Canonical Ensemble (LCE), is an application of canonical ensemble formulation from statistical physics for learning from data. The approach prominently appears in generative artificial intelligence. EBMs provide a unified framework for many probabilistic and non-probabilistic approaches to such learning, particularly for training graphical and other structured models. An EBM learns the characteristics of a target dataset and generates a similar but larger dataset. EBMs detect the latent variables of a dataset and generate new datasets with a similar distribution. Energy-based generative neural networks is a class of generative models, which aim to learn explicit probability distributions of data in the form of energy-based models, the energy functions of which are parameterized by modern deep neural networks. Boltzmann machines are a special form of energy-based models with a specific parametrization of the energy. == Description == For a given input x {\displaystyle x} , the model describes an energy E θ ( x ) {\displaystyle E_{\theta }(x)} such that the Boltzmann distribution P θ ( x ) = e − β E θ ( x ) Z ( θ ) {\displaystyle P_{\theta }(x)={e^{-\beta E_{\theta }(x)} \over Z(\theta )}} is a probability (density), and typically β = 1 {\displaystyle \beta =1} . Since the normalization constant: Z ( θ ) := ∫ x ∈ X e − β E θ ( x ) d x {\displaystyle Z(\theta ):=\int _{x\in X}e^{-\beta E_{\theta }(x)}dx} (also known as the partition function) depends on all the Boltzmann factors of all possible inputs x {\displaystyle x} , it cannot be easily computed or reliably estimated during training simply using standard maximum likelihood estimation. However, for maximizing the likelihood during training, the gradient of the log-likelihood of a single training example x {\displaystyle x} is given by using the chain rule: ∂ θ log ( P θ ( x ) ) = E x ′ ∼ P θ [ ∂ θ E θ ( x ′ ) ] − ∂ θ E θ ( x ) ( ∗ ) {\displaystyle \partial _{\theta }\log \left(P_{\theta }(x)\right)=\mathbb {E} _{x'\sim P_{\theta }}[\partial _{\theta }E_{\theta }(x')]-\partial _{\theta }E_{\theta }(x)\,()} The expectation in the above formula for the gradient can be approximately estimated by drawing samples x ′ {\displaystyle x'} from the distribution P θ {\displaystyle P_{\theta }} using Markov chain Monte Carlo (MCMC). Early energy-based models, such as the 2003 Boltzmann machine by Hinton, estimated this expectation via blocked Gibbs sampling. Newer approaches make use of more efficient Stochastic Gradient Langevin Dynamics (LD), drawing samples using: x 0 ′ ∼ P 0 , x i + 1 ′ = x i ′ − α 2 ∂ E θ ( x i ′ ) ∂ x i ′ + ϵ {\displaystyle x_{0}'\sim P_{0},x_{i+1}'=x_{i}'-{\frac {\alpha }{2}}{\frac {\partial E_{\theta }(x_{i}')}{\partial x_{i}'}}+\epsilon } , where ϵ ∼ N ( 0 , α ) {\displaystyle \epsilon \sim {\mathcal {N}}(0,\alpha )} . A replay buffer of past values x i ′ {\displaystyle x_{i}'} is used with LD to initialize the optimization module. The parameters θ {\displaystyle \theta } of the neural network are therefore trained in a generative manner via MCMC-based maximum likelihood estimation: the learning process follows an "analysis by synthesis" scheme, where within each learning iteration, the algorithm samples the synthesized examples from the current model by a gradient-based MCMC method (e.g., Langevin dynamics or Hybrid Monte Carlo), and then updates the parameters θ {\displaystyle \theta } based on the difference between the training examples and the synthesized ones – see equation ( ∗ ) {\displaystyle ()} . This process can be interpreted as an alternating mode seeking and mode shifting process, and also has an adversarial interpretation. Essentially, the model learns a function E θ {\displaystyle E_{\theta }} that associates low energies to correct values, and higher energies to incorrect values. After training, given a converged energy model E θ {\displaystyle E_{\theta }} , the Metropolis–Hastings algorithm can be used to draw new samples. The acceptance probability is given by: P a c c ( x i → x ∗ ) = min ( 1 , P θ ( x ∗ ) P θ ( x i ) ) . {\displaystyle P_{acc}(x_{i}\to x^{})=\min \left(1,{\frac {P_{\theta }(x^{})}{P_{\theta }(x_{i})}}\right).} == History == The term "energy-based models" was first coined in a 2003 JMLR paper where the authors defined a generalisation of independent components analysis to the overcomplete setting using EBMs. Other early work on EBMs proposed models that represented energy as a composition of latent and observable variables. == Characteristics == EBMs demonstrate useful properties: Simplicity and stability. The EBM is the only object that needs to be designed and trained. Separate networks need not be trained to ensure balance. Adaptive computation time. An EBM can generate sharp, diverse samples or (more quickly) coarse, less diverse samples. Given infinite time, this procedure produces true samples. Flexibility. In Variational Autoencoders (VAE) and flow-based models, the generator learns a map from a continuous space to a (possibly) discontinuous space containing different data modes. EBMs can learn to assign low energies to disjoint regions (multiple modes). Adaptive generation. EBM generators are implicitly defined by the probability distribution, and automatically adapt as the distribution changes (without training), allowing EBMs to address domains where generator training is impractical, as well as minimizing mode collapse and avoiding spurious modes from out-of-distribution samples. Compositionality. Individual models are unnormalized probability distributions, allowing models to be combined through product of experts or other hierarchical techniques. == Experimental results == On image datasets such as CIFAR-10 and ImageNet 32x32, an EBM model generated high-quality images relatively quickly. It supported combining features learned from one type of image for generating other types of images. It was able to generalize using out-of-distribution datasets, outperforming flow-based and autoregressive models. EBM was relatively resistant to adversarial perturbations, behaving better than models explicitly trained against them with training for classification. == Applications == Target applications include natural language processing, robotics and computer vision. The first energy-based generative neural network is the generative ConvNet proposed in 2016 for image patterns, where the neural network is a convolutional neural network. The model has been generalized to various domains to learn distributions of videos, and 3D voxels. They are made more effective in their variants. They have proven useful for data generation (e.g., image synthesis, video synthesis, 3D shape synthesis, etc.), data recovery (e.g., recovering videos with missing pixels or image frames, 3D super-resolution, etc), data reconstruction (e.g., image reconstruction and linear interpolation ). == Alternatives == EBMs compete with techniques such as variational autoencoders (VAEs), generative adversarial networks (GANs) or normalizing flows. == Extensions == === Joint energy-based models === Joint energy-based models (JEM), proposed in 2020 by Grathwohl et al., allow any classifier with softmax output to be interpreted as energy-based model. The key observation is that such a classifier is trained to predict the conditional probability p θ ( y | x ) = e f → θ ( x ) [ y ] ∑ j = 1 K e f → θ ( x ) [ j ] for y = 1 , … , K and f → θ = ( f 1 , … , f K ) ∈ R K , {\displaystyle p_{\theta }(y|x)={\frac {e^{{\vec {f}}_{\theta }(x)[y]}}{\sum _{j=1}^{K}e^{{\vec {f}}_{\theta }(x)[j]}}}\ \ {\text{ for }}y=1,\dotsc ,K{\text{ and }}{\vec {f}}_{\theta }=(f_{1},\dotsc ,f_{K})\in \mathbb {R} ^{K},} where f → θ ( x ) [ y ] {\displaystyle {\vec {f}}_{\theta }(x)[y]} is the y-th index of the logits f → {\displaystyle {\vec {f}}} corresponding to class y. Without any change to the logits it was proposed to reinterpret the logits to describe a joint probability density: p θ ( y , x ) = e f → θ ( x ) [ y ] Z ( θ ) , {\displaystyle p_{\theta }(y,x)={\frac {e^{{\vec {f}}_{\theta }(x)[y]}}{Z(\theta )}},} with unknown partition function Z ( θ ) {\displaystyle Z(\theta )} and energy E θ ( x , y ) = − f θ ( x ) [ y ] {\displaystyle E_{\theta }(x,y)=-f_{\theta }(x)[y]} . By marginalization, we obtain the unnormalized density p θ ( x ) = ∑ y p θ ( y , x ) = ∑ y e f → θ ( x ) [ y ] Z ( θ ) =: e − E θ ( x ) , {\displaystyle p_{\theta }(x)=\sum _{y}p_{\theta }(y,x)=\sum _{y}{\frac {e^{{\vec {f}}_{\theta }(x)[y]}}{Z(\theta )}}=:e^{-E_{\theta }(x)},} therefore, E θ ( x ) = − log ( ∑ y e f → θ ( x ) [ y ] Z ( θ ) ) , {\displaystyle E_{\theta }(x)=-\log \left(\sum _{y}{\frac {e^{{\vec {f}}_{\theta }(x)[y]}}{Z(\theta )}}\right),} so that any classifier can be used to define an energy function E θ ( x ) {\displaystyle E_{\theta }(x)} .
HTTP Strict Transport Security
HTTP Strict Transport Security (HSTS) is a policy mechanism that helps to protect websites against man-in-the-middle attacks such as protocol downgrade attacks and cookie hijacking. It allows web servers to declare that web browsers (or other complying user agents) should automatically interact with it using only HTTPS connections, which provide Transport Layer Security (TLS/SSL), unlike the insecure HTTP used alone. HSTS is an IETF standards track protocol and is specified in RFC 6797. The HSTS Policy is communicated by the server to the user agent via an HTTP response header field named Strict-Transport-Security. HSTS Policy specifies a period of time during which the user agent should only access the server in a secure fashion. Websites using HSTS often do not accept clear text HTTP, either by rejecting connections over HTTP or systematically redirecting users to HTTPS (though this is not required by the specification). The consequence of this is that a user-agent not capable of doing TLS will not be able to connect to the site. The protection normally only applies after a user has visited the site at least once, relying on the principle of "trust on first use". The way this protection works is that when a user entering or selecting an HTTP (not HTTPS) URL to the site, the client, such as a Web browser, will automatically upgrade to HTTPS without making an HTTP request, thereby preventing any HTTP man-in-the-middle attack from occurring. To counteract this problem, an HSTS preload list maintained by Google Chrome and used by other major web browsers is maintained. If a domain is on this list, the browser skips the initial request and encrypts all communication immediately. Additional domains can be registered at no cost. == Specification history == The HSTS specification was published as RFC 6797 on 19 November 2012 after being approved on 2 October 2012 by the IESG for publication as a Proposed Standard RFC. The authors originally submitted it as an Internet Draft on 17 June 2010. With the conversion to an Internet Draft, the specification name was altered from "Strict Transport Security" (STS) to "HTTP Strict Transport Security", because the specification applies only to HTTP. The HTTP response header field defined in the HSTS specification however remains named "Strict-Transport-Security". The last so-called "community version" of the then-named "STS" specification was published on 18 December 2009, with revisions based on community feedback. The original draft specification by Jeff Hodges from PayPal, Collin Jackson, and Adam Barth was published on 18 September 2009. The HSTS specification is based on original work by Jackson and Barth as described in their paper "ForceHTTPS: Protecting High-Security Web Sites from Network Attacks". Additionally, HSTS is the realization of one facet of an overall vision for improving web security, put forward by Jeff Hodges and Andy Steingruebl in their 2010 paper The Need for Coherent Web Security Policy Framework(s). == HSTS mechanism overview == A server implements an HSTS policy by supplying a header over an HTTPS connection (HSTS headers over HTTP are ignored). For example, a server could send a header such that future requests to the domain for the next year (max-age is specified in seconds; 31,536,000 is equal to one non-leap year) use only HTTPS: Strict-Transport-Security: max-age=31536000. When a web application issues HSTS Policy to user agents, conformant user agents behave as follows: Automatically turn any insecure links referencing the web application into secure links (e.g. http://example.com/some/page/ will be modified to https://example.com/some/page/ before accessing the server). If the security of the connection cannot be ensured (e.g. the server's TLS certificate is not trusted), the user agent must terminate the connection and should not allow the user to access the web application. This helps protect web application users against some passive (eavesdropping) and active network attacks. A man-in-the-middle attacker has a greatly reduced ability to intercept requests and responses between a user and a web application server while the user's browser has HSTS Policy in effect for that web application. == Applicability == The most important security vulnerability that HSTS can fix is SSL-stripping man-in-the-middle attacks, first publicly introduced by Moxie Marlinspike in his 2009 BlackHat Federal talk "New Tricks For Defeating SSL In Practice". The SSL (and TLS) stripping attack works by transparently converting a secure HTTPS connection into a plain HTTP connection. The user can see that the connection is insecure, but crucially there is no way of knowing whether the connection should be secure. At the time of Marlinspike's talk, many websites did not use TLS/SSL, therefore there was no way of knowing (without prior knowledge) whether the use of plain HTTP was due to an attack, or simply because the website had not implemented TLS/SSL. Additionally, no warnings are presented to the user during the downgrade process, making the attack fairly subtle to all but the most vigilant. Marlinspike's sslstrip tool, presented at Black Hat DC 2009, fully automates the attack. HSTS addresses this problem by informing the browser that connections to the site should always use TLS/SSL. The HSTS header can be stripped by the attacker if this is the user's first visit. Google Chrome, Mozilla Firefox, Internet Explorer, and Microsoft Edge attempt to limit this problem by including a "pre-loaded" list of HSTS sites. Unfortunately this solution cannot scale to include all websites on the internet. See limitations, below. HSTS can also help to prevent having one's cookie-based website login credentials stolen by widely available tools such as Firesheep. Because HSTS is time limited, it is sensitive to attacks involving shifting the victim's computer time e.g. using false NTP packets. == Limitations == The initial request remains unprotected from active attacks if it uses an insecure protocol such as plain HTTP or if the URI for the initial request was obtained over an insecure channel. The same applies to the first request after the activity period specified in the advertised HSTS Policy max-age (sites should set a period of several days or months depending on user activity and behavior). === Solutions with preload list === Google Chrome, Mozilla Firefox, and Internet Explorer/Microsoft Edge address this limitation by implementing a "HSTS preloaded list", which is a list that contains known sites supporting HSTS. This list is distributed with the browser so that it uses HTTPS for the initial request to the listed sites as well. As previously mentioned, these pre-loaded lists cannot scale to cover the entire Web. A potential solution might be achieved by using DNS records to declare HSTS Policy, and accessing them securely via DNSSEC, optionally with certificate fingerprints to ensure validity (which requires running a validating resolver to avoid last mile issues). Junade Ali has noted that HSTS is ineffective against the use of false domains; by using DNS-based attacks, it is possible for a man-in-the-middle interceptor to serve traffic from an artificial domain which is not on the HSTS Preload list, this can be made possible by DNS Spoofing Attacks, or simply a domain name that misleadingly resembles the real domain name such as www.example.org instead of www.example.com. Even with an HSTS preloaded list, HSTS cannot prevent advanced attacks against TLS itself, such as the BEAST or CRIME attacks introduced by Juliano Rizzo and Thai Duong. Attacks against TLS itself are orthogonal to HSTS policy enforcement. Neither can it protect against attacks on the server - if someone compromises it, it will happily serve any content over TLS. === Privacy issues === HSTS can be used to near-indelibly tag visiting browsers with recoverable identifying data (supercookies) which can persist in and out of browser "incognito" privacy modes. By creating a web page that makes multiple HTTP requests to selected domains, for example, if twenty browser requests to twenty different domains are used, theoretically over one million visitors can be distinguished (220) due to the resulting requests arriving via HTTP vs. HTTPS; the latter being the previously recorded binary "bits" established earlier via HSTS headers. == Browser support == Chromium and Google Chrome since version 4.0.211.0 Firefox since version 4; with Firefox 17, Mozilla integrates a list of websites supporting HSTS. Opera since version 12 Safari since OS X Mavericks (version 10.9, late 2013) Internet Explorer 11 on Windows 8.1 and Windows 7 with KB3058515 installed (Released as a Windows Update in June 2015) Microsoft Edge and Internet Explorer 11 on Windows 10 BlackBerry 10 Browser and WebView since BlackBerry OS 10.3.3. == Deployment best practices == Depending on the actual deployment there are certain threats (e.g. cookie injection attacks) t