An alternating decision tree (ADTree) is a machine learning method for classification. It generalizes decision trees and has connections to boosting. An ADTree consists of an alternation of decision nodes, which specify a predicate condition, and prediction nodes, which contain a single number. An instance is classified by an ADTree by following all paths for which all decision nodes are true, and summing any prediction nodes that are traversed. == History == ADTrees were introduced by Yoav Freund and Llew Mason. However, the algorithm as presented had several typographical errors. Clarifications and optimizations were later presented by Bernhard Pfahringer, Geoffrey Holmes and Richard Kirkby. Implementations are available in Weka and JBoost. == Motivation == Original boosting algorithms typically used either decision stumps or decision trees as weak hypotheses. As an example, boosting decision stumps creates a set of T {\displaystyle T} weighted decision stumps (where T {\displaystyle T} is the number of boosting iterations), which then vote on the final classification according to their weights. Individual decision stumps are weighted according to their ability to classify the data. Boosting a simple learner results in an unstructured set of T {\displaystyle T} hypotheses, making it difficult to infer correlations between attributes. Alternating decision trees introduce structure to the set of hypotheses by requiring that they build off a hypothesis that was produced in an earlier iteration. The resulting set of hypotheses can be visualized in a tree based on the relationship between a hypothesis and its "parent." Another important feature of boosted algorithms is that the data is given a different distribution at each iteration. Instances that are misclassified are given a larger weight while accurately classified instances are given reduced weight. == Alternating decision tree structure == An alternating decision tree consists of decision nodes and prediction nodes. Decision nodes specify a predicate condition. Prediction nodes contain a single number. ADTrees always have prediction nodes as both root and leaves. An instance is classified by an ADTree by following all paths for which all decision nodes are true and summing any prediction nodes that are traversed. This is different from binary classification trees such as CART (Classification and regression tree) or C4.5 in which an instance follows only one path through the tree. === Example === The following tree was constructed using JBoost on the spambase dataset (available from the UCI Machine Learning Repository). In this example, spam is coded as 1 and regular email is coded as −1. The following table contains part of the information for a single instance. The instance is scored by summing all of the prediction nodes through which it passes. In the case of the instance above, the score is calculated as The final score of 0.657 is positive, so the instance is classified as spam. The magnitude of the value is a measure of confidence in the prediction. The original authors list three potential levels of interpretation for the set of attributes identified by an ADTree: Individual nodes can be evaluated for their own predictive ability. Sets of nodes on the same path may be interpreted as having a joint effect The tree can be interpreted as a whole. Care must be taken when interpreting individual nodes as the scores reflect a re weighting of the data in each iteration. == Description of the algorithm == The inputs to the alternating decision tree algorithm are: A set of inputs ( x 1 , y 1 ) , … , ( x m , y m ) {\displaystyle (x_{1},y_{1}),\ldots ,(x_{m},y_{m})} where x i {\displaystyle x_{i}} is a vector of attributes and y i {\displaystyle y_{i}} is either -1 or 1. Inputs are also called instances. A set of weights w i {\displaystyle w_{i}} corresponding to each instance. The fundamental element of the ADTree algorithm is the rule. A single rule consists of a precondition, a condition, and two scores. A condition is a predicate of the form "attribute
Word error rate
Word error rate (WER) is a common metric of the performance of a speech recognition or machine translation system. The WER metric typically ranges from 0 to 1, where 0 indicates that the compared pieces of text are exactly identical, and 1 (or larger) indicates that they are completely different with no similarity. This way, a WER of 0.8 means that there is an 80% error rate for compared sentences. The general difficulty of measuring performance lies in the fact that the recognized word sequence can have a different length from the reference word sequence (supposedly the correct one). The WER is derived from the Levenshtein distance, working at the word level instead of the phoneme level. The WER is a valuable tool for comparing different systems as well as for evaluating improvements within one system. This kind of measurement, however, provides no details on the nature of translation errors and further work is therefore required to identify the main source(s) of error and to focus any research effort. This problem is solved by first aligning the recognized word sequence with the reference (spoken) word sequence using dynamic string alignment. Examination of this issue is seen through a theory called the power law that states the correlation between perplexity and word error rate. Word error rate can then be computed as: W E R = S + D + I N = S + D + I S + D + C {\displaystyle {\mathit {WER}}={\frac {S+D+I}{N}}={\frac {S+D+I}{S+D+C}}} where S is the number of substitutions, D is the number of deletions, I is the number of insertions, C is the number of correct words, N is the number of words in the reference (N=S+D+C) The intuition behind 'deletion' and 'insertion' is how to get from the reference to the hypothesis. So if we have the reference "This is wikipedia" and hypothesis "This _ wikipedia", we call it a deletion. Note that since N is the number of words in the reference, the word error rate can be larger than 1.0, namely if the number of insertions I is larger than the number of correct words C. When reporting the performance of a speech recognition system, sometimes word accuracy (WAcc) is used instead: W A c c = 1 − W E R = N − S − D − I N = C − I N {\displaystyle {\mathit {WAcc}}=1-{\mathit {WER}}={\frac {N-S-D-I}{N}}={\frac {C-I}{N}}} Since the WER can be larger than 1.0, the word accuracy can be smaller than 0.0. == Experiments == It is commonly believed that a lower word error rate shows superior accuracy in recognition of speech, compared with a higher word error rate. However, at least one study has shown that this may not be true. In a Microsoft Research experiment, it was shown that, if people were trained under "that matches the optimization objective for understanding", (Wang, Acero and Chelba, 2003) they would show a higher accuracy in understanding of language than other people who demonstrated a lower word error rate, showing that true understanding of spoken language relies on more than just high word recognition accuracy. == Other metrics == One problem with using a generic formula such as the one above, however, is that no account is taken of the effect that different types of error may have on the likelihood of successful outcome, e.g. some errors may be more disruptive than others and some may be corrected more easily than others. These factors are likely to be specific to the syntax being tested. A further problem is that, even with the best alignment, the formula cannot distinguish a substitution error from a combined deletion plus insertion error. Hunt (1990) has proposed the use of a weighted measure of performance accuracy where errors of substitution are weighted at unity but errors of deletion and insertion are both weighted only at 0.5, thus: W E R = S + 0.5 D + 0.5 I N {\displaystyle {\mathit {WER}}={\frac {S+0.5D+0.5I}{N}}} There is some debate, however, as to whether Hunt's formula may properly be used to assess the performance of a single system, as it was developed as a means of comparing more fairly competing candidate systems. A further complication is added by whether a given syntax allows for error correction and, if it does, how easy that process is for the user. There is thus some merit to the argument that performance metrics should be developed to suit the particular system being measured. Whichever metric is used, however, one major theoretical problem in assessing the performance of a system is deciding whether a word has been “mis-pronounced,” i.e. does the fault lie with the user or with the recogniser. This may be particularly relevant in a system which is designed to cope with non-native speakers of a given language or with strong regional accents. The pace at which words should be spoken during the measurement process is also a source of variability between subjects, as is the need for subjects to rest or take a breath. All such factors may need to be controlled in some way. For text dictation it is generally agreed that performance accuracy at a rate below 95% is not acceptable, but this again may be syntax and/or domain specific, e.g. whether there is time pressure on users to complete the task, whether there are alternative methods of completion, and so on. The term "Single Word Error Rate" is sometimes referred to as the percentage of incorrect recognitions for each different word in the system vocabulary. == Edit distance == The word error rate may also be referred to as the length normalized edit distance. The normalized edit distance between X and Y, d( X, Y ) is defined as the minimum of W( P ) / L ( P ), where P is an editing path between X and Y, W ( P ) is the sum of the weights of the elementary edit operations of P, and L(P) is the number of these operations (length of P).
Memory-hard function
In cryptography, a memory-hard function (MHF) is a function that costs a significant amount of memory to efficiently evaluate. It differs from a memory-bound function, which incurs cost by slowing down computation through memory latency. MHFs have found use in key stretching and proof of work as their increased memory requirements significantly reduce the computational efficiency advantage of custom hardware over general-purpose hardware compared to non-MHFs. == Introduction == MHFs are designed to consume large amounts of memory on a computer in order to reduce the effectiveness of parallel computing. In order to evaluate the function using less memory, a significant time penalty is incurred. As each MHF computation requires a large amount of memory, the number of function computations that can occur simultaneously is limited by the amount of available memory. This reduces the efficiency of specialised hardware, such as application-specific integrated circuits and graphics processing units, which utilise parallelisation, in computing a MHF for a large number of inputs, such as when brute-forcing password hashes or mining cryptocurrency. == Motivation and examples == Bitcoin's proof-of-work uses repeated evaluation of the SHA-256 function, but modern general-purpose processors, such as off-the-shelf CPUs, are inefficient when computing a fixed function many times over. Specialized hardware, such as application-specific integrated circuits (ASICs) designed for Bitcoin mining, can use 30,000 times less energy per hash than x86 CPUs whilst having much greater hash rates. This led to concerns about the centralization of mining for Bitcoin and other cryptocurrencies. Because of this inequality between miners using ASICs and miners using CPUs or off-the shelf hardware, designers of later proof-of-work systems utilised hash functions for which it was difficult to construct ASICs that could evaluate the hash function significantly faster than a CPU. As memory cost is platform-independent, MHFs have found use in cryptocurrency mining, such as for Litecoin, which uses scrypt as its hash function. They are also useful in password hashing because they significantly increase the cost of trying many possible passwords against a leaked database of hashed passwords without significantly increasing the computation time for legitimate users. == Measuring memory hardness == There are various ways to measure the memory hardness of a function. One commonly seen measure is cumulative memory complexity (CMC). In a parallel model, CMC is the sum of the memory required to compute a function over every time step of the computation. Other viable measures include integrating memory usage against time and measuring memory bandwidth consumption on a memory bus. Functions requiring high memory bandwidth are sometimes referred to as "bandwidth-hard functions". == Variants == MHFs can be categorized into two different groups based on their evaluation patterns: data-dependent memory-hard functions (dMHF) and data-independent memory-hard functions (iMHF). As opposed to iMHFs, the memory access pattern of a dMHF depends on the function input, such as the password provided to a key derivation function. Examples of dMHFs are scrypt and Argon2d, while examples of iMHFs are Argon2i and catena. Many of these MHFs have been designed to be used as password hashing functions because of their memory hardness. A notable problem with dMHFs is that they are prone to side-channel attacks such as cache timing. This has resulted in a preference for using iMHFs when hashing passwords. However, iMHFs have been mathematically proven to have weaker memory hardness properties than dMHFs.
Multiple encryption
Multiple encryption is the process of encrypting an already encrypted message one or more times, either using the same or a different algorithm. It is also known as cascade encryption, cascade ciphering, cipher stacking, multiple encryption, and superencipherment. Superencryption refers to the outer-level encryption of a multiple encryption. Some cryptographers, like Matthew Green of Johns Hopkins University, say multiple encryption addresses a problem that mostly doesn't exist: Modern ciphers rarely get broken... You’re far more likely to get hit by malware or an implementation bug than you are to suffer a catastrophic attack on Advanced Encryption Standard (AES). However, from the previous quote an argument for multiple encryption can be made, namely poor implementation. Using two different cryptomodules and keying processes from two different vendors requires both vendors' wares to be compromised for security to fail completely. == Independent keys == Picking any two ciphers, if the key used is the same for both, the second cipher could possibly undo the first cipher, partly or entirely. This is true of ciphers where the decryption process is exactly the same as the encryption process (a reciprocal cipher) – the second cipher would completely undo the first. If an attacker were to recover the key through cryptanalysis of the first encryption layer, the attacker could possibly decrypt all the remaining layers, assuming the same key is used for all layers. To prevent that risk, one can use keys that are statistically independent for each layer (e.g. independent RNGs). Ideally each key should have separate and different generation, sharing, and management processes. == Independent Initialization Vectors == For en/decryption processes that require sharing an Initialization Vector (IV) / nonce these are typically, openly shared or made known to the recipient (and everyone else). Its good security policy never to provide the same data in both plaintext and ciphertext when using the same key and IV. Therefore, its recommended (although at this moment without specific evidence) to use separate IVs for each layer of encryption. == Importance of the first layer == With the exception of the one-time pad, no cipher has been theoretically proven to be unbreakable. Furthermore, some recurring properties may be found in the ciphertexts generated by the first cipher. Since those ciphertexts are the plaintexts used by the second cipher, the second cipher may be rendered vulnerable to attacks based on known plaintext properties (see references below). This is the case when the first layer is a program P that always adds the same string S of characters at the beginning (or end) of all ciphertexts (commonly known as a magic number). When found in a file, the string S allows an operating system to know that the program P has to be launched in order to decrypt the file. This string should be removed before adding a second layer. To prevent this kind of attack, one can use the method provided by Bruce Schneier: Generate a random pad R of the same size as the plaintext. Encrypt R using the first cipher and key. XOR the plaintext with the pad, then encrypt the result using the second cipher and a different (!) key. Concatenate both ciphertexts in order to build the final ciphertext. A cryptanalyst must break both ciphers to get any information. This will, however, have the drawback of making the ciphertext twice as long as the original plaintext. Note, however, that a weak first cipher may merely make a second cipher that is vulnerable to a chosen plaintext attack also vulnerable to a known plaintext attack. However, a block cipher must not be vulnerable to a chosen plaintext attack to be considered secure. Therefore, the second cipher described above is not secure under that definition, either. Consequently, both ciphers still need to be broken. The attack illustrates why strong assumptions are made about secure block ciphers and ciphers that are even partially broken should never be used. == The Rule of Two == The Rule of Two is a data security principle from the NSA's Commercial Solutions for Classified Program (CSfC). It specifies two completely independent layers of cryptography to protect data. For example, data could be protected by both hardware encryption at its lowest level and software encryption at the application layer. It could mean using two FIPS-validated software cryptomodules from different vendors to en/decrypt data. The importance of vendor and/or model diversity between the layers of components centers around removing the possibility that the manufacturers or models will share a vulnerability. This way if one components is compromised there is still an entire layer of encryption protecting the information at rest or in transit. The CSfC Program offers solutions to achieve diversity in two ways. "The first is to implement each layer using components produced by different manufacturers. The second is to use components from the same manufacturer, where that manufacturer has provided NSA with sufficient evidence that the implementations of the two components are independent of one another." The principle is practiced in the NSA's secure mobile phone called Fishbowl. The phones use two layers of encryption protocols, IPsec and Secure Real-time Transport Protocol (SRTP), to protect voice communications. The Samsung Galaxy S9 Tactical Edition is also an approved CSfC Component.
Cryptographic High Value Product
Cryptographic High Value Product (CHVP) is a designation used within the information security community to identify assets that have high value, and which may be used to encrypt / decrypt secure communications, but which do not retain or store any classified information. When disconnected from the secure communication network, the CHVP equipment may be handled with a lower level of controls than required for COMSEC equipment.
Distributed concurrency control
Distributed concurrency control is the concurrency control of a system distributed over a computer network (Bernstein et al. 1987, Weikum and Vossen 2001). In database systems and transaction processing (transaction management) distributed concurrency control refers primarily to the concurrency control of a distributed database. It also refers to the concurrency control in a multidatabase (and other multi-transactional object) environment (e.g., federated database, grid computing, and cloud computing environments. A major goal for distributed concurrency control is distributed serializability (or global serializability for multidatabase systems). Distributed concurrency control poses special challenges beyond centralized one, primarily due to communication and computer latency. It often requires special techniques, like distributed lock manager over fast computer networks with low latency, like switched fabric (e.g., InfiniBand). The most common distributed concurrency control technique is strong strict two-phase locking (SS2PL, also named rigorousness), which is also a common centralized concurrency control technique. SS2PL provides both the serializability and strictness. Strictness, a special case of recoverability, is utilized for effective recovery from failure. For large-scale distribution and complex transactions, distributed locking's typical heavy performance penalty (due to delays, latency) can be saved by using the atomic commitment protocol, which is needed in a distributed database for (distributed) transactions' atomicity.
Big memory
Big-memory computers are machines with a large amount of random-access memory (RAM). The computers are required for databases, graph analytics, or more generally, high-performance computing, data science, and big data. Some database systems called in-memory databases are designed to run mostly in memory, rarely if ever retrieving data from disk or flash memory. See list of in-memory databases. == Details == The performance of big-memory systems depends on how the central processing units (CPUs) access the memory, via a conventional memory controller or via non-uniform memory access (NUMA). Performance also depends on the size and design of the CPU cache. Performance also depends on operating system (OS) design. The huge pages feature in Linux and other OSes can improve the efficiency of virtual memory. The transparent huge pages feature in Linux can offer better performance for some big-memory workloads. The "Large-Page Support" in Microsoft Windows enables server applications to establish large-page memory regions which are typically three orders of magnitude larger than the native page size.