AI Chatbot Development

AI Chatbot Development — independent reviews, comparisons, pricing and step-by-step guides on Aizhi.

Frankenstein complex

The Frankenstein complex is a term coined by Isaac Asimov in his robot series, referring to the fear of mechanical men. == History == Some of Asimov's science fiction short stories and novels predict that this suspicion will become strongest and most widespread in respect of "mechanical men" that most-closely resemble human beings (see android), but it is also present on a lower level against robots that are plainly electromechanical automatons. The "Frankenstein complex" is similar in many respects to Masahiro Mori's uncanny valley hypothesis. The name, "Frankenstein complex", is derived from the name of Victor Frankenstein in the 1818 novel Frankenstein; or, The Modern Prometheus by Mary Shelley. In Shelley's story, Frankenstein created an intelligent, somewhat superhuman being, but he finds that his creation is horrifying to behold and abandons it. This ultimately leads to Victor's death at the conclusion of a vendetta between himself and his creation. In much of his fiction, Asimov depicts the general attitude of the public towards robots as negative, with ordinary people fearing that robots will either replace them or dominate them, although dominance would not be allowed under the specifications of the Three Laws of Robotics, the first of which is: "A robot may not harm a human being or, through inaction, allow a human being to come to harm." However, Asimov's fictitious earthly public is not fully persuaded by this, and remains largely suspicious and fearful of robots. I, Robot's short story "Little Lost Robot" is about this "fear of robots". In Asimov's robot novels, the Frankenstein complex is a major problem for roboticists and robot manufacturers. They do all they can to reassure the public that robots are harmless, even though this sometimes involves hiding the truth because they think that the public would misunderstand it. The fear by the public and the response of the manufacturers is an example of the theme of paternalism, the dread of paternalism, and the conflicts that arise from it in Asimov's fiction. The same theme occurs in many later works of fiction featuring robots, although it is rarely referred to as such.
Read more →
Top 10 AI Voice Assistants Compared (2026)

Comparing the best AI voice assistant? An AI voice assistant is software that uses machine learning to help you get more done — it lowers the barrier so anyone can produce professional output. Privacy matters too: check whether your data trains the model and whether a no-log or enterprise tier is available. Whether you are a beginner or a pro, the right AI voice assistant slots into your workflow and pays for itself fast. Below we compare features, pricing, and real output so you can choose with confidence.
Read more →
Sparse dictionary learning

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 d {\displaystyle n>d} with the latter being a typical assumption for a sparse dictionary learning problem. The case of a complete dictionary does not provide any improvement from a representational point of view and thus isn't considered. Undercomplete dictionaries represent the setup in which the actual input data lies in a lower-dimensional space. This case is strongly related to dimensionality reduction and techniques like principal component analysis which require atoms d 1 , . . . , d n {\displaystyle d_{1},...,d_{n}} to be orthogonal. The choice of these subspaces is crucial for efficient dimensionality reduction, but it is not trivial. And dimensionality reduction based on dictionary representation can be extended to address specific tasks such as data analysis or classification. However, their main downside is limiting the choice of atoms. Overcomplete dictionaries, however, do not require the atoms to be orthogonal (they will never have a basis anyway) thus allowing for more flexible dictionaries and richer data representations. An overcomplete dictionary which allows for sparse representation of signal can be a famous transform matrix (wavelets transform, fourier transform) or it can be formulated so that its elements are changed in such a way that it sparsely represents the given signal in a best way. Learned dictionaries are capable of giving sparser solutions as compared to predefined transform matrices. == Algorithms == As the optimization problem described above can be solved as a convex problem with respect to either dictionary or sparse coding while the other one of the two is fixed, most of the algorithms are based on the idea of iteratively updating one and then the other. The problem of finding an optimal sparse coding R {\displaystyle R} with a given dictionary D {\displaystyle \mathbf {D} } is known as sparse approximation (or sometimes just sparse coding problem). A number of algorithms have been developed to solve it (such as matching pursuit and LASSO) and are incorporated in the algorithms described below. === Method of optimal directions (MOD) === The method of optimal directions (or MOD) was one of the first methods introduced to tackle the sparse dictionary learning problem. The core idea of it is to solve the minimization problem subject to the limited number of non-zero components of the representation vector: min D , R { ‖ X − D R ‖ F 2 } s.t. ∀ i ‖ r i ‖ 0 ≤ T {\displaystyle \min _{\mathbf {D} ,R}\{\|X-\mathbf {D} R\|_{F}^{2}\}\,\,{\text{s.t.}}\,\,\forall i\,\,\|r_{i}\|_{0}\leq T} Here, F {\displaystyle F} denotes the Frobenius norm. MOD alternates between getting the sparse coding using a method such as matching pursuit and updating the dictionary by computing the analytical solution of the problem given by D = X R + {\displaystyle \mathbf {D} =XR^{+}} where R + {\displaystyle R^{+}} is a Moore-Penrose pseudoinverse. After this update D {\displaystyle \mathbf {D} } is renormalized to fit the constraints and the new sparse coding is obtained again. The process is repeated until convergence (or until a sufficiently small residue). MOD has proved to be a very efficient method for low-dimensional input data X {\displaystyle X} requiring just a few iterations to converge. However, due to the high complexity of the matrix-inversion operation, computing the pseudoinverse in high-dimensional cases is in many cases intractable. This shortcoming has inspired the development of other dictionary learning methods. === K-SVD === K-SVD is an algorithm that performs SVD at its core to update the atoms of the dictionary one by one and basically is a generalization of K-means. It enforces that each element of the input data x i {\displaystyle x_{i}} is encoded by a linear combination of not more than T 0 {\displaystyle T_{0}} elements in a way identical to the MOD approach: min D , R { ‖ X − D R ‖ F 2 } s.t. ∀ i ‖ r i ‖ 0 ≤ T 0 {\displaystyle \min _{\mathbf {D} ,R}\{\|X-\mathbf {D} R\|_{F}^{2}\}\,\,{\text{s.t.}}\,\,\forall i\,\,\|r_{i}\|_{0}\leq T_{0}} This algorithm's essence is to first fix the dictionary, find the best possible R {\displaystyle R} under the above constraint (using Orthogonal Matching Pursuit) and then iteratively update the atoms of dictionary D {\displaystyle \mathbf {D} } in the following manner: ‖ X − D R ‖ F 2 = | X − ∑ i = 1 K d i x T i | F 2 = ‖ E k − d k x T k ‖ F 2 {\displaystyle \|X-\mathbf {D} R\|_{F}^{2}=\left|X-\sum _{i=1}^{K}d_{i}x_{T}^{i}\right|_{F}^{2}=\|E_{k}-d_{k}x_{T}^{k}\|_{F}^{2}} The next steps of the algorithm include rank-1 approximation of the residual matrix E k {\displaystyle E_{k}} , updating d k {\displaystyle d_{k}} and enforcing the s
Read more →
Sumio Watanabe

Sumio Watanabe (渡辺澄夫, Watanabe Sumio; born 1959) is a Japanese mathematician and engineer working in probability theory, applied algebraic geometry and Bayesian statistics. He is currently a professor at Tokyo Institute of Technology in the Department of Computational Intelligence and Systems Science. He is the author of the text, Algebraic Geometry and Statistical Learning Theory, which proposes a generalization of Fisher's regular statistical theory to singular statistical models. == Books == Mathematical Theory of Bayesian Statistics, CRC Press, 2018, ISBN 9781482238068 Algebraic Geometry and Statistical Learning Theory, Cambridge University Press, 2009.
Read more →
Conditional random field

Conditional random fields (CRFs) are a class of statistical modeling methods often applied in pattern recognition and machine learning and used for structured prediction. Whereas a classifier predicts a label for a single sample without considering "neighbouring" samples, a CRF can take context into account. To do so, the predictions are modelled as a graphical model, which represents the presence of dependencies between the predictions. The kind of graph used depends on the application. For example, in natural language processing, "linear chain" CRFs are popular, for which each prediction is dependent only on its immediate neighbours. In image processing, the graph typically connects locations to nearby and/or similar locations to enforce that they receive similar predictions. Other examples where CRFs are used are: labeling or parsing of sequential data for natural language processing or biological sequences, part-of-speech tagging, shallow parsing, named entity recognition, gene finding, peptide critical functional region finding, and object recognition and image segmentation in computer vision. == Description == CRFs are a type of discriminative undirected probabilistic graphical model. Lafferty, McCallum and Pereira define a CRF on observations X {\displaystyle {\boldsymbol {X}}} and random variables Y {\displaystyle {\boldsymbol {Y}}} as follows: Let G = ( V , E ) {\displaystyle G=(V,E)} be a graph such that Y = ( Y v ) v ∈ V {\displaystyle {\boldsymbol {Y}}=({\boldsymbol {Y}}_{v})_{v\in V}} , so that Y {\displaystyle {\boldsymbol {Y}}} is indexed by the vertices of G {\displaystyle G} . Then ( X , Y ) {\displaystyle ({\boldsymbol {X}},{\boldsymbol {Y}})} is a conditional random field when each random variable Y v {\displaystyle {\boldsymbol {Y}}_{v}} , conditioned on X {\displaystyle {\boldsymbol {X}}} , obeys the Markov property with respect to the graph; that is, its probability is dependent only on its neighbours in G and not its past states: P ( Y v | X , { Y w : w ≠ v } ) = P ( Y v | X , { Y w : w ∼ v } ) {\displaystyle P({\boldsymbol {Y}}_{v}|{\boldsymbol {X}},\{{\boldsymbol {Y}}_{w}:w\neq v\})=P({\boldsymbol {Y}}_{v}|{\boldsymbol {X}},\{{\boldsymbol {Y}}_{w}:w\sim v\})} , where w ∼ v {\displaystyle {\mathit {w}}\sim v} means that w {\displaystyle w} and v {\displaystyle v} are neighbors in G {\displaystyle G} . What this means is that a CRF is an undirected graphical model whose nodes can be divided into exactly two disjoint sets X {\displaystyle {\boldsymbol {X}}} and Y {\displaystyle {\boldsymbol {Y}}} , the observed and output variables, respectively; the conditional distribution p ( Y | X ) {\displaystyle p({\boldsymbol {Y}}|{\boldsymbol {X}})} is then modeled. === Inference === For general graphs, the problem of exact inference in CRFs is intractable. The inference problem for a CRF is basically the same as for an MRF and the same arguments hold. However, there exist special cases for which exact inference is feasible: If the graph is a chain or a tree, message passing algorithms yield exact solutions. The algorithms used in these cases are analogous to the forward-backward and Viterbi algorithm for the case of HMMs. If the CRF only contains pair-wise potentials and the energy is submodular, combinatorial min cut/max flow algorithms yield exact solutions. If exact inference is impossible, several algorithms can be used to obtain approximate solutions. These include: Loopy belief propagation Alpha expansion Mean field inference Linear programming relaxations === Parameter learning === Learning the parameters θ {\displaystyle \theta } is usually done by maximum likelihood learning for p ( Y i | X i ; θ ) {\displaystyle p(Y_{i}|X_{i};\theta )} . If all nodes have exponential family distributions and all nodes are observed during training, this optimization is convex. It can be solved for example using gradient descent algorithms, or Quasi-Newton methods such as the L-BFGS algorithm. On the other hand, if some variables are unobserved, the inference problem has to be solved for these variables. Exact inference is intractable in general graphs, so approximations have to be used. === Examples === In sequence modeling, the graph of interest is usually a chain graph. An input sequence of observed variables X {\displaystyle X} represents a sequence of observations and Y {\displaystyle Y} represents a hidden (or unknown) state variable that needs to be inferred given the observations. The Y i {\displaystyle Y_{i}} are structured to form a chain, with an edge between each Y i − 1 {\displaystyle Y_{i-1}} and Y i {\displaystyle Y_{i}} . As well as having a simple interpretation of the Y i {\displaystyle Y_{i}} as "labels" for each element in the input sequence, this layout admits efficient algorithms for: model training, learning the conditional distributions between the Y i {\displaystyle Y_{i}} and feature functions from some corpus of training data. decoding, determining the probability of a given label sequence Y {\displaystyle Y} given X {\displaystyle X} . inference, determining the most likely label sequence Y {\displaystyle Y} given X {\displaystyle X} . The conditional dependency of each Y i {\displaystyle Y_{i}} on X {\displaystyle X} is defined through a fixed set of feature functions of the form f ( i , Y i − 1 , Y i , X ) {\displaystyle f(i,Y_{i-1},Y_{i},X)} , which can be thought of as measurements on the input sequence that partially determine the likelihood of each possible value for Y i {\displaystyle Y_{i}} . The model assigns each feature a numerical weight and combines them to determine the probability of a certain value for Y i {\displaystyle Y_{i}} . Linear-chain CRFs have many of the same applications as conceptually simpler hidden Markov models (HMMs), but relax certain assumptions about the input and output sequence distributions. An HMM can loosely be understood as a CRF with very specific feature functions that use constant probabilities to model state transitions and emissions. Conversely, a CRF can loosely be understood as a generalization of an HMM that makes the constant transition probabilities into arbitrary functions that vary across the positions in the sequence of hidden states, depending on the input sequence. Notably, in contrast to HMMs, CRFs can contain any number of feature functions, the feature functions can inspect the entire input sequence X {\displaystyle X} at any point during inference, and the range of the feature functions need not have a probabilistic interpretation. == Variants == === Higher-order CRFs and semi-Markov CRFs === CRFs can be extended into higher order models by making each Y i {\displaystyle Y_{i}} dependent on a fixed number k {\displaystyle k} of previous variables Y i − k , . . . , Y i − 1 {\displaystyle Y_{i-k},...,Y_{i-1}} . In conventional formulations of higher order CRFs, training and inference are only practical for small values of k {\displaystyle k} (such as k ≤ 5), since their computational cost increases exponentially with k {\displaystyle k} . However, another recent advance has managed to ameliorate these issues by leveraging concepts and tools from the field of Bayesian nonparametrics. Specifically, the CRF-infinity approach constitutes a CRF-type model that is capable of learning infinitely-long temporal dynamics in a scalable fashion. This is effected by introducing a novel potential function for CRFs that is based on the Sequence Memoizer (SM), a nonparametric Bayesian model for learning infinitely-long dynamics in sequential observations. To render such a model computationally tractable, CRF-infinity employs a mean-field approximation of the postulated novel potential functions (which are driven by an SM). This allows for devising efficient approximate training and inference algorithms for the model, without undermining its capability to capture and model temporal dependencies of arbitrary length. There exists another generalization of CRFs, the semi-Markov conditional random field (semi-CRF), which models variable-length segmentations of the label sequence Y {\displaystyle Y} . This provides much of the power of higher-order CRFs to model long-range dependencies of the Y i {\displaystyle Y_{i}} , at a reasonable computational cost. Finally, large-margin models for structured prediction, such as the structured Support Vector Machine can be seen as an alternative training procedure to CRFs. === Latent-dynamic conditional random field === Latent-dynamic conditional random fields (LDCRF) or discriminative probabilistic latent variable models (DPLVM) are a type of CRFs for sequence tagging tasks. They are latent variable models that are trained discriminatively. In an LDCRF, like in any sequence tagging task, given a sequence of observations x = x 1 , … , x n {\displaystyle x_{1},\dots ,x_{n}} , the main problem the model must solve is how to assign a sequence of labels y = y 1 , … , y n {\displaystyle y_{1},\dots ,y_{n}} from one finite set
Read more →
Powerset construction

In the theory of computation and automata theory, the powerset construction or subset construction is a standard method for converting a nondeterministic finite automaton (NFA) into a deterministic finite automaton (DFA) that recognizes the same formal language. It is important in theory because it establishes that NFAs, despite their additional flexibility, are unable to recognize any language that cannot be recognized by some DFA. It is also important in practice for converting easier-to-construct NFAs into more efficiently executable DFAs. However, if the NFA has n states, the resulting DFA may have up to 2n states, an exponentially larger number, which sometimes makes the construction impractical for large NFAs. The construction, sometimes called the Rabin–Scott powerset construction (or subset construction) to distinguish it from similar constructions for other types of automata, was first published by Michael O. Rabin and Dana Scott in 1959. == Intuition == To simulate the operation of a DFA on a given input string, one needs to keep track of a single state at any time: the state that the automaton will reach after seeing a prefix of the input. In contrast, to simulate an NFA, one needs to keep track of a set of states: all of the states that the automaton could reach after seeing the same prefix of the input, according to the nondeterministic choices made by the automaton. If, after a certain prefix of the input, a set S of states can be reached, then after the next input symbol x the set of reachable states is a deterministic function of S and x. Therefore, the sets of reachable NFA states play the same role in the NFA simulation as single DFA states play in the DFA simulation, and in fact the sets of NFA states appearing in this simulation may be re-interpreted as being states of a DFA. == Construction == The powerset construction applies most directly to an NFA that does not allow state transformations without consuming input symbols (aka: "ε-moves"). Such an automaton may be defined as a 5-tuple (Q, Σ, T, q0, F), in which Q is the set of states, Σ is the set of input symbols, T is the transition function (mapping a state and an input symbol to a set of states), q0 is the initial state, and F is the set of accepting states. The corresponding DFA has states corresponding to subsets of Q. The initial state of the DFA is {q0}, the (one-element) set of initial states. The transition function of the DFA maps a state S (representing a subset of Q) and an input symbol x to the set T(S,x) = ∪{T(q,x) | q ∈ S}, the set of all states that can be reached by an x-transition from a state in S. A state S of the DFA is an accepting state if and only if at least one member of S is an accepting state of the NFA. In the simplest version of the powerset construction, the set of all states of the DFA is the powerset of Q, the set of all possible subsets of Q. However, many states of the resulting DFA may be useless as they may be unreachable from the initial state. An alternative version of the construction creates only the states that are actually reachable. === NFA with ε-moves === For an NFA with ε-moves (also called an ε-NFA), the construction must be modified to deal with these by computing the ε-closure of states: the set of all states reachable from some given state using only ε-moves. Van Noord recognizes three possible ways of incorporating this closure computation in the powerset construction: Compute the ε-closure of the entire automaton as a preprocessing step, producing an equivalent NFA without ε-moves, then apply the regular powerset construction. This version, also discussed by Hopcroft and Ullman, is straightforward to implement, but impractical for automata with large numbers of ε-moves, as commonly arise in natural language processing application. During the powerset computation, compute the ε-closure { q ′ | q → ε ∗ q ′ } {\displaystyle \{q'~|~q\to _{\varepsilon }^{}q'\}} of each state q that is considered by the algorithm (and cache the result). During the powerset computation, compute the ε-closure { q ′ | ∃ q ∈ Q ′ , q → ε ∗ q ′ } {\displaystyle \{q'~|~\exists q\in Q',q\to _{\varepsilon }^{}q'\}} of each subset of states Q' that is considered by the algorithm, and add its elements to Q'. === Multiple initial states === If NFAs are defined to allow for multiple initial states, the initial state of the corresponding DFA is the set of all initial states of the NFA, or (if the NFA also has ε-moves) the set of all states reachable from initial states by ε-moves. == Example == The NFA below has four states; state 1 is initial, and states 3 and 4 are accepting. Its alphabet consists of the two symbols 0 and 1, and it has ε-moves. The initial state of the DFA constructed from this NFA is the set of all NFA states that are reachable from state 1 by ε-moves; that is, it is the set {1,2,3}. A transition from {1,2,3} by input symbol 0 must follow either the arrow from state 1 to state 2, or the arrow from state 3 to state 4. Additionally, neither state 2 nor state 4 have outgoing ε-moves. Therefore, T({1,2,3},0) = {2,4}, and by the same reasoning the full DFA constructed from the NFA is as shown below. As can be seen in this example, there are five states reachable from the start state of the DFA; the remaining 11 sets in the powerset of the set of NFA states are not reachable. == Complexity == Because the DFA states consist of sets of NFA states, an n-state NFA may be converted to a DFA with at most 2n states. For every n, there exist n-state NFAs such that every subset of states is reachable from the initial subset, so that the converted DFA has exactly 2n states, giving Θ(2n) worst-case time complexity. A simple example requiring nearly this many states is the language of strings over the alphabet {0,1} in which there are at least n characters, the nth from last of which is 1. It can be represented by an (n + 1)-state NFA, but it requires 2n DFA states, one for each n-character suffix of the input; cf. picture for n=4. == Applications == Brzozowski's algorithm for DFA minimization uses the powerset construction, twice. It converts the input DFA into an NFA for the reverse language, by reversing all its arrows and exchanging the roles of initial and accepting states, converts the NFA back into a DFA using the powerset construction, and then repeats its process. Its worst-case complexity is exponential, unlike some other known DFA minimization algorithms, but in many examples it performs more quickly than its worst-case complexity would suggest. Safra's construction, which converts a non-deterministic Büchi automaton with n states into a deterministic Muller automaton or into a deterministic Rabin automaton with 2O(n log n) states, uses the powerset construction as part of its machinery.
Read more →
Two-way finite automaton

In computer science, in particular in automata theory, a two-way finite automaton is a finite automaton that is allowed to re-read its input. == Two-way deterministic finite automaton == A two-way deterministic finite automaton (2DFA) is an abstract machine, a generalized version of the deterministic finite automaton (DFA) which can revisit characters already processed. As in a DFA, there are a finite number of states with transitions between them based on the current character, but each transition is also labelled with a value indicating whether the machine will move its position in the input to the left, right, or stay at the same position. Equivalently, 2DFAs can be seen as read-only Turing machines with no work tape, only a read-only input tape. 2DFAs were introduced in a seminal 1959 paper by Rabin and Scott, who proved them to have equivalent power to one-way DFAs. That is, any formal language which can be recognized by a 2DFA can be recognized by a DFA which only examines and consumes each character in order. Since DFAs are obviously a special case of 2DFAs, this implies that both kinds of machines recognize precisely the class of regular languages. However, the equivalent DFA for a 2DFA may require exponentially many states, making 2DFAs a much more practical representation for algorithms for some common problems. 2DFAs are also equivalent to read-only Turing machines that use only a constant amount of space on their work tape, since any constant amount of information can be incorporated into the finite control state via a product construction (a state for each combination of work tape state and control state). == Formal description == Formally, a two-way deterministic finite automaton can be described by the following 8-tuple: M = ( Q , Σ , L , R , δ , s , t , r ) {\displaystyle M=(Q,\Sigma ,L,R,\delta ,s,t,r)} where Q {\displaystyle Q} is the finite, non-empty set of states Σ {\displaystyle \Sigma } is the finite, non-empty set of input symbols L {\displaystyle L} is the left endmarker R {\displaystyle R} is the right endmarker δ : Q × ( Σ ∪ { L , R } ) → Q × { l e f t , r i g h t } {\displaystyle \delta :Q\times (\Sigma \cup \{L,R\})\rightarrow Q\times \{\mathrm {left,right} \}} s {\displaystyle s} is the start state t {\displaystyle t} is the end state r {\displaystyle r} is the reject state In addition, the following two conditions must also be satisfied: For all q ∈ Q {\displaystyle q\in Q} δ ( q , L ) = ( q ′ , r i g h t ) {\displaystyle \delta (q,L)=(q^{\prime },\mathrm {right} )} for some q ′ ∈ Q {\displaystyle q^{\prime }\in Q} δ ( q , R ) = ( q ′ , l e f t ) {\displaystyle \delta (q,R)=(q^{\prime },\mathrm {left} )} for some q ′ ∈ Q {\displaystyle q^{\prime }\in Q} It says that there must be some transition possible when the pointer reaches either end of the input word. For all symbols σ ∈ Σ ∪ { L } {\displaystyle \sigma \in \Sigma \cup \{L\}} δ ( t , σ ) = ( t , R ) {\displaystyle \delta (t,\sigma )=(t,R)} δ ( r , σ ) = ( r , R ) {\displaystyle \delta (r,\sigma )=(r,R)} δ ( t , R ) = ( t , L ) {\displaystyle \delta (t,R)=(t,L)} δ ( r , R ) = ( r , L ) {\displaystyle \delta (r,R)=(r,L)} It says that once the automaton reaches the accept or reject state, it stays in there forever and the pointer goes to the right most symbol and cycles there infinitely. == Two-way nondeterministic finite automaton == A two-way nondeterministic finite automaton (2NFA) may have multiple transitions defined in the same configuration. Its transition function is δ : Q × ( Σ ∪ { L , R } ) → 2 Q × { l e f t , r i g h t } {\displaystyle \delta :Q\times (\Sigma \cup \{L,R\})\rightarrow 2^{Q\times \{\mathrm {left,right} \}}} . Like a standard one-way NFA, a 2NFA accepts a string if at least one of the possible computations is accepting. Like the 2DFAs, the 2NFAs also accept only regular languages. == Two-way alternating finite automaton == A two-way alternating finite automaton (2AFA) is a two-way extension of an alternating finite automaton (AFA). Its state set is Q = Q ∃ ∪ Q ∀ {\displaystyle Q=Q_{\exists }\cup Q_{\forall }} where Q ∃ ∩ Q ∀ = ∅ {\displaystyle Q_{\exists }\cap Q_{\forall }=\emptyset } . States in Q ∃ {\displaystyle Q_{\exists }} and Q ∀ {\displaystyle Q_{\forall }} are called existential resp. universal. In an existential state a 2AFA nondeterministically chooses the next state like an NFA, and accepts if at least one of the resulting computations accepts. In a universal state 2AFA moves to all next states, and accepts if all the resulting computations accept. == State complexity tradeoffs == Two-way and one-way finite automata, deterministic and nondeterministic and alternating, accept the same class of regular languages. However, transforming an automaton of one type to an equivalent automaton of another type incurs a blow-up in the number of states. Christos Kapoutsis determined that transforming an n {\displaystyle n} -state 2DFA to an equivalent DFA requires n ( n n − ( n − 1 ) n ) {\displaystyle n(n^{n}-(n-1)^{n})} states in the worst case. If an n {\displaystyle n} -state 2DFA or a 2NFA is transformed to an NFA, the worst-case number of states required is ( 2 n n + 1 ) = O ( 4 n n ) {\displaystyle {\binom {2n}{n+1}}=O\left({\frac {4^{n}}{\sqrt {n}}}\right)} . Ladner, Lipton and Stockmeyer. proved that an n {\displaystyle n} -state 2AFA can be converted to a DFA with 2 n 2 n {\displaystyle 2^{n2^{n}}} states. The 2AFA to NFA conversion requires 2 Θ ( n log ⁡ n ) {\displaystyle 2^{\Theta (n\log n)}} states in the worst case, see Geffert and Okhotin. It is an open problem whether every 2NFA can be converted to a 2DFA with only a polynomial increase in the number of states. The problem was raised by Sakoda and Sipser, who compared it to the P vs. NP problem in the computational complexity theory. Berman and Lingas discovered a formal relation between this problem and the L vs. NL open problem, see Kapoutsis for a precise relation. == Sweeping automata == Sweeping automata are 2DFAs of a special kind that process the input string by making alternating left-to-right and right-to-left sweeps, turning only at the endmarkers. Sipser constructed a sequence of languages, each accepted by an n-state NFA, yet which is not accepted by any sweeping automata with fewer than 2 n {\displaystyle 2^{n}} states. == Two-way quantum finite automaton == The concept of 2DFAs was in 1997 generalized to quantum computing by John Watrous's "On the Power of 2-Way Quantum Finite State Automata", in which he demonstrates that these machines can recognize nonregular languages and so are more powerful than DFAs. == Two-way pushdown automaton == A pushdown automaton that is allowed to move either way on its input tape is called two-way pushdown automaton (2PDA); it has been studied by Hartmanis, Lewis, and Stearns (1965). Aho, Hopcroft, Ullman (1968) and Cook (1971) characterized the class of languages recognizable by deterministic (2DPDA) and non-deterministic (2NPDA) two-way pushdown automata; Gray, Harrison, and Ibarra (1967) investigated the closure properties of these languages.
Read more →
StarDict

StarDict, developed by Hu Zheng (胡正), is a free GUI released under the GPL-3.0-or-later license for accessing StarDict dictionary files (a dictionary shell). It is the successor of StarDic, developed by Ma Su'an (馬蘇安), continuing its version numbers. According to StarDict's earlier homepage on SourceForge, the project has been removed from SourceForge due to copyright infringement reports. It moved to Google Code and then back to SourceForge, while development is now seemingly continued on GitHub. == Supported platforms == StarDict runs under Linux, Windows, FreeBSD, Maemo and Solaris. Dictionaries of the user's choice are installed separately. Dictionary files can be created by converting dict files. Several programs compatible with the StarDict dictionary format are available for different platforms. For the iPhone, iPod Touch and iPad, applications available in the App Store include GuruDic, TouchDict, weDict, Dictionary Universal, Alpus and others, as well as the free iStarDict, which is available for the Cydia Store. == Dictionaries available == One can find here the partial list of FreeDict dictionaries which can be converted to the StarDict format. These include, in particular, some older versions of Webster's dictionary and many dictionaries for various languages. == Features == While StarDict is in scan mode, results are displayed in a tooltip, allowing easy dictionary lookup. When combined with Freedict, StarDict will quickly provide rough translations of foreign language websites. On September 25, 2006, an online version of Stardict began operation. This online version includes access to all the major dictionaries of StarDict, as well as Wikipedia in Chinese. Previous versions of StarDict were very similar to the PowerWord dictionary program, which is developed by a Chinese company, KingSoft. Since version 2.4.2, however, StarDict has diverged from the design of PowerWord by increasing its search capabilities and adding lexicons in a variety of languages. This was assisted by the collaboration of many developers with the author. == sdcv == Evgeniy A. Dushistov produced a command line version of StarDict called sdcv. It employed all the dictionary files that belong to StarDict. It is written in C++ and licensed under the terms of the GNU General Public License. sdcv runs under Linux, FreeBSD, and Solaris. As in StarDict, dictionaries of the user's choice have to be installed separately. At the end of 2006, software developer Hu Zheng cited personal financial problems as an excuse to charge users for downloading dictionary files from his website, which temporarily aroused strong doubts and dissatisfaction in the Linux community. In the end, under the pressure of public opinion, the charging plan was forced to be canceled and ended hastily.
Read more →
NNDB

The Notable Names Database (NNDB) is an online database of biographical details of over 40,000 people. Soylent Communications, a sole proprietorship that also hosted the later defunct Rotten.com, describes NNDB as an "intelligence aggregator" of noteworthy persons, highlighting their interpersonal connections. The Rotten.com domain was registered in 1996 by former Apple and Netscape software engineer Thomas E. Dell, who was also known by his internet alias, "Soylent". == Entries == Each entry has an executive summary followed by a brief narrative about their life. It also lists date and cause of death if deceased. Businesspeople and government officials are listed with chronologies of their posts, positions, and board memberships. As of 2022, the site is no longer updated. == NNDB Mapper == The NNDB Mapper, a visual tool for exploring connections between people, was made available in May 2008. It required Adobe Flash 7.
Read more →
Cognitive computer

A cognitive computer is a computer that hardwires artificial intelligence and machine learning algorithms into an integrated circuit that closely reproduces the behavior of the human brain. It generally adopts a neuromorphic engineering approach. Synonyms include neuromorphic chip and cognitive chip. In 2023, IBM's proof-of-concept NorthPole chip (optimized for 2-, 4- and 8-bit precision) achieved remarkable performance in image recognition. In 2013, IBM developed Watson, a cognitive computer that uses neural networks and deep learning techniques. The following year, it developed the 2014 TrueNorth microchip architecture which is designed to be closer in structure to the human brain than the von Neumann architecture used in conventional computers. In 2017, Intel also announced its version of a cognitive chip in "Loihi, which it intended to be available to university and research labs in 2018. Intel (most notably with its Pohoiki Beach and Springs systems), Qualcomm, and others are improving neuromorphic processors steadily. == IBM TrueNorth chip == TrueNorth was a neuromorphic CMOS integrated circuit produced by IBM in 2014. It is a manycore processor network on a chip design, with 4096 cores, each one having 256 programmable simulated neurons for a total of just over a million neurons. In turn, each neuron has 256 programmable "synapses" that convey the signals between them. Hence, the total number of programmable synapses is just over 268 million (228). Its basic transistor count is 5.4 billion. In 2023 Zhejiang University and Alibaba developed Darwin a neuromorphic chip The darwin3 chip was designed around 2023 so it is fairly modern compared to IBM's TrueNorth or Intel's LoihI. === Details === Memory, computation, and communication are handled in each of the 4096 neurosynaptic cores, TrueNorth circumvents the von Neumann-architecture bottleneck and is very energy-efficient, with IBM claiming a power consumption of 70 milliwatts and a power density that is 1/10,000th of conventional microprocessors. The SyNAPSE chip operates at lower temperatures and power because it only draws power necessary for computation. Skyrmions have been proposed as models of the synapse on a chip. The neurons are emulated using a Linear-Leak Integrate-and-Fire (LLIF) model, a simplification of the leaky integrate-and-fire model. According to IBM, it does not have a clock, operates on unary numbers, and computes by counting to a maximum of 19 bits. The cores are event-driven by using both synchronous and asynchronous logic, and are interconnected through an asynchronous packet-switched mesh network on chip (NOC). IBM developed a new network to program and use TrueNorth. It included a simulator, a new programming language, an integrated programming environment, and libraries. This lack of backward compatibility with any previous technology (e.g., C++ compilers) poses serious vendor lock-in risks and other adverse consequences that may prevent it from commercialization in the future. === Research === In 2018, a cluster of TrueNorth network-linked to a master computer was used in stereo vision research that attempted to extract the depth of rapidly moving objects in a scene. == IBM NorthPole chip == In 2023, IBM released its NorthPole chip, which is a proof-of-concept for dramatically improving performance by intertwining compute with memory on-chip, thus eliminating the Von Neumann bottleneck. It blends approaches from IBM's 2014 TrueNorth system with modern hardware designs to achieve speeds about 4,000 times faster than TrueNorth. It can run ResNet-50 or Yolo-v4 image recognition tasks about 22 times faster, with 25 times less energy and 5 times less space, when compared to GPUs which use the same 12-nm node process that it was fabricated with. It includes 224 MB of RAM and 256 processor cores and can perform 2,048 operations per core per cycle at 8-bit precision, and 8,192 operations at 2-bit precision. It runs at between 25 and 425 MHz. This is an inferencing chip, but it cannot yet handle GPT-4 because of memory and accuracy limitations == Intel Loihi chip == === Pohoiki Springs === Pohoiki Springs is a system that incorporates Intel's self-learning neuromorphic chip, named Loihi, introduced in 2017, perhaps named after the Hawaiian seamount Lōʻihi. Intel claims Loihi is about 1000 times more energy efficient than general-purpose computing systems used to train neural networks. In theory, Loihi supports both machine learning training and inference on the same silicon independently of a cloud connection, and more efficiently than convolutional neural networks or deep learning neural networks. Intel points to a system for monitoring a person's heartbeat, taking readings after events such as exercise or eating, and using the chip to normalize the data and work out the ‘normal’ heartbeat. It can then spot abnormalities and deal with new events or conditions. The first iteration of the chip was made using Intel's 14 nm fabrication process and houses 128 clusters of 1,024 artificial neurons each for a total of 131,072 simulated neurons. This offers around 130 million synapses, far less than the human brain's 800 trillion synapses, and behind IBM's TrueNorth. Loihi is available for research purposes among more than 40 academic research groups as a USB form factor. In October 2019, researchers from Rutgers University published a research paper to demonstrate the energy efficiency of Intel's Loihi in solving simultaneous localization and mapping. In March 2020, Intel and Cornell University published a research paper to demonstrate the ability of Intel's Loihi to recognize different hazardous materials, which could eventually aid to "diagnose diseases, detect weapons and explosives, find narcotics, and spot signs of smoke and carbon monoxide". === Pohoiki Beach === Intel's Loihi 2, named Pohoiki Beach, was released in September 2021 with 64 cores. It boasts faster speeds, higher-bandwidth inter-chip communications for enhanced scalability, increased capacity per chip, a more compact size due to process scaling, and improved programmability. === Hala Point === Hala Point packages 1,152 Loihi 2 processors produced on Intel 3 process node in a six-rack-unit chassis. The system supports up to 1.15 billion neurons and 128 billion synapses distributed over 140,544 neuromorphic processing cores, consuming 2,600 watts of power. It includes over 2,300 embedded x86 processors for ancillary computations. Intel claimed in 2024 that Hala Point was the world’s largest neuromorphic system. It uses Loihi 2 chips. It is claimed to offer 10x more neuron capacity and up to 12x higher performance. The Darwin3 chip exceeds these specs. Hala Point provides up to 20 quadrillion operations per second, (20 petaops), with efficiency exceeding 15 trillion (8-bit) operations s−1 W−1 on conventional deep neural networks. Hala Point integrates processing, memory and communication channels in a massively parallelized fabric, providing 16 PB s−1 of memory bandwidth, 3.5 PB s−1 of inter-core communication bandwidth, and 5 TB s−1 of inter-chip bandwidth. The system can process its 1.15 billion neurons 20 times faster than a human brain. Its neuron capacity is roughly equivalent to that of an owl brain or the cortex of a capuchin monkey. Loihi-based systems can perform inference and optimization using 100 times less energy at speeds as much as 50 times faster than CPU/GPU architectures. Intel claims that Hala Point can create LLMs. Much further research is needed == SpiNNaker == SpiNNaker (Spiking Neural Network Architecture) is a massively parallel, manycore supercomputer architecture designed by the Advanced Processor Technologies Research Group at the Department of Computer Science, University of Manchester. == Criticism == Critics argue that a room-sized computer – as in the case of IBM's Watson – is not a viable alternative to a three-pound human brain. Some also cite the difficulty for a single system to bring so many elements together, such as the disparate sources of information as well as computing resources. In 2021, The New York Times released Steve Lohr's article "What Ever Happened to IBM’s Watson?". He wrote about some costly failures of IBM Watson. One of them, a cancer-related project called the Oncology Expert Advisor, was abandoned in 2016 as a costly failure. During the collaboration, Watson could not use patient data. Watson struggled to decipher doctors’ notes and patient histories. The development of LLMs has placed a new emphasis on cognitive computers, because the Transformer technology that underpins LLMs demands huge energy for GPUs and PCs. Cognitive computers use significantly less energy, but the details of STDPs and neuron models cannot yet match the accuracy of backprop, and so ANN to SNN weight translations such as QAT and PQT or progressive quantization are becoming popular, with their own limitations.
Read more →
Extended affix grammar

In computer science, extended affix grammars (EAGs) are a formal grammar formalism for describing the context free and context sensitive syntax of language, both natural language and programming languages. EAGs are a member of the family of two-level grammars; more specifically, a restriction of Van Wijngaarden grammars with the specific purpose of making parsing feasible. Like Van Wijngaarden grammars, EAGs have hyperrules that form a context-free grammar except in that their nonterminals may have arguments, known as affixes, the possible values of which are supplied by another context-free grammar, the metarules. EAGs were introduced and studied by D.A. Watt in 1974; recognizers were developed at the University of Nijmegen between 1985 and 1995. The EAG compiler developed there will generate either a recogniser, a transducer, a translator, or a syntax directed editor for a language described in the EAG formalism. The formalism is quite similar to Prolog, to the extent that it borrowed its cut operator. EAGs have been used to write grammars of natural languages such as English, Spanish, and Hungarian. The aim was to verify the grammars by making them parse corpora of text (corpus linguistics); hence, parsing had to be sufficiently practical. However, the parse tree explosion problem that ambiguities in natural language tend to produce in this type of approach is worsened for EAGs because each choice of affix value may produce a separate parse, even when several different values are equivalent. The remedy proposed was to switch to the much simpler Affix Grammar over a Finite Lattice (AGFL) instead, in which metagrammars can only produce simple finite languages.
Read more →
Postediting

Post-editing (or postediting) is the process whereby humans amend machine-generated translation to achieve an acceptable final product. A person who post-edits is called a post-editor. The concept of post-editing is linked to that of pre-editing. In the process of translating a text via machine translation, best results may be gained by pre-editing the source text – for example by applying the principles of controlled language – and then post-editing the machine output. It is distinct from editing, which refers to the process of improving human generated text (a process which is often known as revision in the field of translation). Post-edited text may afterwards be revised to ensure the quality of the language choices are proofread to correct simple mistakes. Post-editing involves the correction of machine translation output to ensure that it meets a level of quality negotiated in advance between the client and the post-editor. Light post-editing aims at making the output simply understandable; full post-editing at making it also stylistically appropriate. With advances in machine translation full post-editing is becoming an alternative to manual translation. Practically all computer-assisted translation (CAT) tools now support post-editing of machine translated output. == Post-editing and machine translation == Machine translation left the labs to start being used for its actual purpose in the late seventies at some big institutions such as the European Commission and the Pan-American Health Organization, and then, later, at some corporations such as Caterpillar and General Motors. First studies on post-editing appeared in the eighties, linked to those implementations. To develop appropriate guidelines and training, members of the Association for Machine Translation in the Americas (AMTA) and the European Association for Machine Translation (EAMT) set a Post-editing Special Interest Group in 1999. After the nineties, advances in computer power and connectivity sped machine translation development and allowed for its deployment through the web browser, including as a free, useful adjunct to the main search engines (Google Translate, Bing Translator, Yahoo! Babel Fish). A wider acceptance of less than perfect machine translation was accompanied also by a wider acceptance of post-editing. With the demand for localisation of goods and services growing at a pace that could not be met by human translation, not even assisted by translation memory and other translation management technologies, industry bodies such as the Translation Automation Users Society (TAUS) expect machine translation and post-editing to play a much bigger role within the next few years. The use of Machine Translation suggests sometimes pre-editing. Human translators possess significantly more sophisticated cognitive abilities than machine translation (MT) systems. They leverage a wealth of life experience, common sense, and multi-sensory input to understand context, identify semantic intent, and add cultural nuances to translations. This remains true even as MT capabilities continue to improve. Unlike MT systems, which primarily focus on literal word-for-word conversion, human translators grasp the underlying meaning and intent, even when information is implicit. They "read between the lines," guided by their understanding of the world. Essentially, MT models excel at text string prediction, not true comprehension. Their success often stems from framing problems as prediction tasks, such as in self-driving cars or fraud detection. Studies have demonstrated that integrating adaptive MT with post-editing interfaces can lead to reductions in technical effort and time, improving overall translation efficiency. These systems are also supported by research that highlights the benefits of adaptive MT in real-world translation scenarios. For example, incremental adaptation in Neural Machine Translation (NMT) for professional post-editors has been shown to improve translation quality and reduce time spent on edits, showcasing how human expertise and machine assistance can complement each other effectively. == Light and full post-editing == For many years, no widely accepted, standardized post-editing guidelines existed; however, in 2017, ISO standard 18587:2017: Translation services — Post-editing of machine translation output — Requirements was published. Studies in the eighties distinguished between degrees of post-editing which, in the context of the European Commission Translation Service, were first defined as conventional and rapid or full and rapid. Light and full post-editing seems the wording most used today. Light post-editing implies minimal intervention by the post-editor, with the aim of ensuring quality is "good enough" or "understandable"; the expectation is that the client will use it for inbound purposes only, often when the text is needed urgently, or has a short time span. Full post-editing involves a greater level of intervention to achieve a degree of quality to be negotiated between client and post-editor; the expectation is that the outcome will be a text that is not only understandable but presented in some stylistically appropriate way, so it can be used for assimilation and even for dissemination, for inbound and for outbound purposes. The quality is expected to be publishable and equivalent to that of a human translation. The assumption, however, has been that it takes less effort for translators to work directly from the source text than to post-edit the machine generated version. With advances in machine translation, this may be changing. For some language pairs and for some tasks, and with engines that have been customised with domain specific good quality data, some clients are already requesting translators to post-edit instead of translating from scratch, in the belief that they will attain similar quality at a lower cost. The light/full classification, developed in the nineties when machine translation still came on a CD-ROM, may not suit advances in machine translation at the light post-editing end either. For some language pairs and some tasks, particularly if the source has been pre-edited, raw machine output may be good enough for gisting purposes without requiring subsequent human intervention. == Post-editing efficiency == Post-editing is used when raw machine translation is not good enough and human translation not required. Industry advises post-editing to be used when it can at least double the productivity of manual translation, even fourfold it in the case of light post-editing (1000 words per hour vs. 250 wph). However, post-editing efficiency is difficult to predict. Various studies from both academia and industry have claimed that post-editing is generally faster than translating from scratch, regardless of language pairs or translators' experience. There is, however, no agreement about how much time can be saved through post-editing in practice (if any at all): While the industry reports on time savings around 40%, some academic studies suggest that time savings under actual working conditions are more likely to be between 0–20%, or that it may depend on the terminological proximity between the source and target languages. Professionals have also reported negative productivity gains where corrections require more time than to translate from scratch. == Post-editing and the language industry == After some thirty years, post-editing is still "a nascent profession". What the right profile of the post-editor is, has not yet been fully studied. Post-editing overlaps with translating and editing, but only partially. Most think the ideal post-editor will be a translator keen to be trained on the specific skills required, but there are some who think a bilingual without a background in translation may be easier to train. Not much is known either on who the actual post-editors are, whether they tend to be professional translators, whether they work mostly as in-house employees or self-employed, and on which conditions. Many professional translators dislike post-editing, among other reasons because it tends to be paid at lower rates than conventional translations, with the International Association of Professional Translators and Interpreters (IAPTI) having been particularly vocal about it.
Read more →
Random (software)

Random was an iOS mobile app that used algorithms and human-curation to create an adaptive interface to the Internet. The app served a remix of relevance and serendipity that allowed people to find diverse topics and interesting content that they might not have encountered otherwise. Random did not require a login or sign-up - the use of the app was anonymous. The app was powered by an artificial intelligence that learned from direct and indirect user interactions inside the app. While learning and adapting to a person, Random created a unique anonymous choice profile that was then used for recommending topics and content. The app didn't recommend the same content twice. == User interface == Random's user interface was made of ever-changing topic blocks that contained keywords and images. By choosing any of the blocks, the user would see related web content. By closing the web content, the user could access new related topics. The user interface allowed people to get more information about a specific topic area or then just leap freely from topic to topic. The content recommended by Random could be any type of web content, varying from news articles to long-form stories and from photographs to videos. Every user of the Random was curating content for other users by using the app. == History == Random was launched in March 2014. The startup was backed by Skype co-founder Janus Friis. The Random app received a strong reception from the likes of The New York Times, TechCrunch, New Scientist, Vice, and other leading publications. The app went on to gain traction with an active and loyal user community of several hundreds of thousands. This was not enough to support the free app model the team strongly believed in, and the service was terminated in December 2015. == Reception == Various reviews in media have emphasized that Random enables people to break their filter bubble and find diverse content they might not find elsewhere. Alan Henry of Lifehacker wrote: "Random... breaks you out by intentionally guiding you to new topics and interesting articles at sites you may not otherwise read." Vice Motherboard's Claire Evans says that: "Random never turns into a filter bubble, because it perpetually injects the irrational into my experience… in a cocktail of relevancy and serendipity." The app has been said to have a unique, minimalistic user experience. Kit Eaton of The New York Times commented that Random "let's you browse the news in a different way to all the other news sites you've probably ever used." Mashable reviewed Random by concluding that the "app may be one of the most simple content-discovery apps on the market."
Read more →
Nicholas Carlini

Nicholas Carlini is an American researcher affiliated with Anthropic and previously with Google DeepMind who has published research in the fields of computer security and machine learning. He is known for his work on adversarial machine learning, particularly his work on the Carlini & Wagner attack in 2016. This attack was particularly useful in defeating defensive distillation, a method used to increase model robustness, and has since been effective against other defenses against adversarial input. In 2018, Carlini demonstrated an attack on Mozilla's DeepSpeech model, showing that hidden commands could be embedded in speech inputs, which the model would execute even if they were inaudible to humans. He also led a team at UC Berkeley that successfully broke seven out of nine defenses against adversarial attacks presented at the 2018 International Conference on Learning Representations. In addition to his work on adversarial attacks, Carlini has made significant contributions to understanding the privacy risks of machine learning models. In 2020, he revealed that large language models, like GPT-2, could memorize and output personally identifiable information. His research demonstrated that this issue worsened with larger models, and he later showed similar vulnerabilities in generative image models, such as Stable Diffusion. == Life and career == Nicholas Carlini obtained his Bachelor of Arts in Computer Science and Mathematics from the University of California, Berkeley, in 2013. He then continued his studies at the same university, where he pursued a PhD under the supervision of David Wagner, completing it in 2018. Carlini became known for his work on adversarial machine learning. In 2016, he worked alongside Wagner to develop the Carlini & Wagner attack, a method of generating adversarial examples against machine learning models. The attack was proved to be useful against defensive distillation, a popular mechanism where a student model is trained based on the features of a parent model to increase the robustness and generalizability of student models. The attack gained popularity when it was shown that the methodology was also effective against most other defenses, rendering them ineffective. In 2018, Carlini demonstrated an attack against Mozilla Foundation's DeepSpeech model where he showed that by hiding malicious commands inside normal speech input the speech model would respond to the hidden commands even when the commands were not discernible by humans. In the same year, Carlini and his team at UC Berkeley showed that out of the 11 papers presenting defenses to adversarial attacks accepted in that year's ICLR conference, seven of the defenses could be broken. Since 2021, he and his team have been working on large language models, creating a questionnaire where humans typically scored 35% whereas AI models scored in the 40%, with GPT-3 getting 38% which could be improved to 40% through few shot prompting. The best performer in the test was UnifiedQA, a model developed by Google specifically for answer questions and answer sets. Carlini has also developed methods to cause large language models like ChatGPT to answer harmful questions like how to construct bombs. He is also known for his work studying the privacy of machine learning models. In 2020, he showed for the first time that large language models would memorize some of the text data that they were trained on. For example, he found that GPT-2 could output personally identifiable information. He then led an analysis of larger models and studied how memorization increased with model size. Then, in 2022 he showed the same vulnerability in generative image models, and specifically diffusion models, by showing that Stable Diffusion could output images of people's faces that it was trained on. Following on this, Carlini then showed that ChatGPT would also sometimes output exact copies of webpages it was trained on, including personally identifiable information. Some of these studies have since been referenced by the courts in debating the copyright status of AI models. == Other work == Carlini received the Best of Show award at the 2020 IOCCC for implementing a tic-tac-toe game entirely with calls to printf, expanding on work from a research paper of his from 2015. The judges commented on his submission "This year's Best of Show (carlini) is such a novel way of obfuscation that it would be worth of a special mention in the (future) Best of IOCCC list!". [sic] == Awards == Best Student Paper Award, IEEE S&P 2017 ("Towards Evaluating the Robustness of Neural Networks") Best Paper Award, ICML 2018 ("Obfuscated Gradients Give a False Sense of Security: Circumventing Defenses to Adversarial Examples") Distinguished Paper Award, USENIX 2021 ("Poisoning the Unlabeled Dataset of Semi-Supervised Learning") Distinguished Paper Award, USENIX 2023 ("Tight Auditing of Differentially Private Machine Learning") Best Paper Award, ICML 2024 ("Stealing Part of a Production Language Model") Best Paper Award, ICML 2024 ("Considerations for Differentially Private Learning with Large-Scale Public Pretraining")
Read more →
Localization Industry Standards Association

Localization Industry Standards Association or LISA was a Swiss-based trade body concerning the translation of computer software (and associated materials) into multiple natural languages, which existed from 1990 to February 2011. It counted among its members most of the large information technology companies of the period, including Adobe, Cisco, Hewlett-Packard, IBM, McAfee, Nokia, Novell and Xerox. LISA played a significant role in representing its partners at the International Organization for Standardization (ISO), and the TermBase eXchange (TBX) standard developed by LISA was submitted to ISO in 2007 and became ISO 30042:2008. LISA also had a presence at the W3C. A number of the LISA standards are used by the OASIS Open Architecture for XML Authoring and Localization framework. LISA shut down on 28 February 2011, and its website went offline shortly afterwards. In the wake of the closure of LISA, the European Telecommunications Standards Institute started an Industry Specification Group (ISG) for localization. The ISG has five work items: Term-Base eXchange (TBX) / ISO 30042:2008 Translation Memory eXchange (TMX), with GALA Segmentation Rules eXchange (SRX) / ISO/CD 24621) Global information management Metrics eXchange – Volume (GMX-V); Another organization that was formed in response to the closure of LISA is Terminology for Large Organizations (TerminOrgs), a consortium of terminology professionals who promote terminology management best practices.
Read more →