In computer science, the Knuth–Eve algorithm is an algorithm for polynomial evaluation. It preprocesses the coefficients of the polynomial to reduce the number of multiplications required at runtime. Ideas used in the algorithm were originally proposed by Donald Knuth in 1962. His procedure opportunistically exploits structure in the polynomial being evaluated. In 1964, James Eve determined for which polynomials this structure exists, and gave a simple method of "preconditioning" polynomials (explained below) to endow them with that structure. == Algorithm == === Preliminaries === Consider an arbitrary polynomial p ∈ R [ x ] {\displaystyle p\in \mathbb {R} [x]} of degree n {\displaystyle n} . Assume that n ≥ 3 {\displaystyle n\geq 3} . Define m {\displaystyle m} such that: if n {\displaystyle n} is odd then n = 2 m + 1 {\displaystyle n=2m+1} , and if n {\displaystyle n} is even then n = 2 m + 2 {\displaystyle n=2m+2} . Unless otherwise stated, all variables in this article represent either real numbers or univariate polynomials with real coefficients. All operations in this article are done over R {\displaystyle \mathbb {R} } . Again, the goal is to create an algorithm that returns p ( x ) {\displaystyle p(x)} given any x {\displaystyle x} . The algorithm is allowed to depend on the polynomial p {\displaystyle p} itself, since its coefficients are known in advance. === Overview === ==== Key idea ==== Using polynomial long division, we can write p ( x ) = q ( x ) ⋅ ( x 2 − α ) + ( β x + γ ) , {\displaystyle p(x)=q(x)\cdot (x^{2}-\alpha )+(\beta x+\gamma ),} where x 2 − α {\displaystyle x^{2}-\alpha } is the divisor. Picking a value for α {\displaystyle \alpha } fixes both the quotient q {\displaystyle q} and the coefficients in the remainder β {\displaystyle \beta } and γ {\displaystyle \gamma } . The key idea is to cleverly choose α {\displaystyle \alpha } such that β = 0 {\displaystyle \beta =0} , so that p ( x ) = q ( x ) ⋅ ( x 2 − α ) + γ . {\displaystyle p(x)=q(x)\cdot (x^{2}-\alpha )+\gamma .} This way, no operations are needed to compute the remainder polynomial, since it's just a constant. We apply this procedure recursively to q {\displaystyle q} , expressing p ( x ) = ( ( q ( x ) ⋅ ( x 2 − α m ) + γ m ) ⋯ ) ⋅ ( x 2 − α 1 ) + γ 1 . {\displaystyle p(x)=\left(\left(q(x)\cdot (x^{2}-\alpha _{m})+\gamma _{m}\right)\cdots \right)\cdot (x^{2}-\alpha _{1})+\gamma _{1}.} After m {\displaystyle m} recursive calls, the quotient q {\displaystyle q} is either a linear or a quadratic polynomial. In this base case, the polynomial can be evaluated with (say) Horner's method. ==== "Preconditioning" ==== For arbitrary p {\displaystyle p} , it may not be possible to force β = 0 {\displaystyle \beta =0} at every step of the recursion. Consider the polynomials p e {\displaystyle p^{e}} and p o {\displaystyle p^{o}} with coefficients taken from the even and odd terms of p {\displaystyle p} respectively, so that p ( x ) = p e ( x 2 ) + x ⋅ p o ( x 2 ) . {\displaystyle p(x)=p^{e}(x^{2})+x\cdot p^{o}(x^{2}).} If every root of p o {\displaystyle p^{o}} is real, then it is possible to write p {\displaystyle p} in the form given above. Each α i {\displaystyle \alpha _{i}} is a different root of p o {\displaystyle p^{o}} , counting multiple roots as distinct. Furthermore, if at least n − 1 {\displaystyle n-1} roots of p {\displaystyle p} lie in one half of the complex plane, then every root of p o {\displaystyle p^{o}} is real. Ultimately, it may be necessary to "precondition" p {\displaystyle p} by shifting it — by setting p ( x ) ← p ( x + t ) {\displaystyle p(x)\gets p(x+t)} for some t {\displaystyle t} — to endow it with the structure that most of its roots lie in one half of the complex plane. At runtime, this shift has to be "undone" by first setting x ← x − t {\displaystyle x\gets x-t} . === Preprocessing step === The following algorithm is run once for a given polynomial p {\displaystyle p} . At this point, the values of x {\displaystyle x} that p {\displaystyle p} will be evaluated on are not known. ==== Better choice of t ==== While any t ≥ Re ( r 2 ) {\displaystyle t\geq {\text{Re}}(r_{2})} can work, it is possible to remove one addition during evaluation if t {\displaystyle t} is also chosen such that two roots of p ( x + t ) {\displaystyle p(x+t)} are symmetric about the origin. In that case, α 1 {\displaystyle \alpha _{1}} can be chosen such that the shifted polynomial has a factor of x 2 − α 1 {\displaystyle x^{2}-\alpha _{1}} , so γ 1 = 0 {\displaystyle \gamma _{1}=0} . It is always possible to find such a t {\displaystyle t} . One possible algorithm for choosing t {\displaystyle t} is: === Evaluation step === The following algorithm evaluates p {\displaystyle p} at some, now known, point x {\displaystyle x} . Assuming t {\displaystyle t} is chosen optimally, γ 1 = 0 {\displaystyle \gamma _{1}=0} . So, the final iteration of the loop can instead run y ← y ⋅ ( s − α i ) , {\displaystyle y\gets y\cdot (s-\alpha _{i}),} saving an addition. == Analysis == In total, evaluation using the Knuth–Eve algorithm for a polynomial of degree n {\displaystyle n} requires n {\displaystyle n} additions and ⌊ n / 2 ⌋ + 2 {\displaystyle \lfloor n/2\rfloor +2} multiplications, assuming t {\displaystyle t} is chosen optimally. No algorithm to evaluate a given polynomial of degree n {\displaystyle n} can use fewer than n {\displaystyle n} additions or fewer than ⌈ n / 2 ⌉ {\displaystyle \lceil n/2\rceil } multiplications during evaluation. This result assumes only addition and multiplication are allowed during both preprocessing and evaluation. The Knuth–Eve algorithm is not well-conditioned.
Learnable function class
In statistical learning theory, a learnable function class is a set of functions for which an algorithm can be devised to asymptotically minimize the expected risk, uniformly over all probability distributions. The concept of learnable classes are closely related to regularization in machine learning, and provides large sample justifications for certain learning algorithms. == Definition == === Background === Let Ω = X × Y = { ( x , y ) } {\displaystyle \Omega ={\mathcal {X}}\times {\mathcal {Y}}=\{(x,y)\}} be the sample space, where y {\displaystyle y} are the labels and x {\displaystyle x} are the covariates (predictors). F = { f : X ↦ Y } {\displaystyle {\mathcal {F}}=\{f:{\mathcal {X}}\mapsto {\mathcal {Y}}\}} is a collection of mappings (functions) under consideration to link x {\displaystyle x} to y {\displaystyle y} . L : Y × Y ↦ R {\displaystyle L:{\mathcal {Y}}\times {\mathcal {Y}}\mapsto \mathbb {R} } is a pre-given loss function (usually non-negative). Given a probability distribution P ( x , y ) {\displaystyle P(x,y)} on Ω {\displaystyle \Omega } , define the expected risk I P ( f ) {\displaystyle I_{P}(f)} to be: I P ( f ) = ∫ L ( f ( x ) , y ) d P ( x , y ) {\displaystyle I_{P}(f)=\int L(f(x),y)dP(x,y)} The general goal in statistical learning is to find the function in F {\displaystyle {\mathcal {F}}} that minimizes the expected risk. That is, to find solutions to the following problem: f ^ = arg min f ∈ F I P ( f ) {\displaystyle {\hat {f}}=\arg \min _{f\in {\mathcal {F}}}I_{P}(f)} But in practice the distribution P {\displaystyle P} is unknown, and any learning task can only be based on finite samples. Thus we seek instead to find an algorithm that asymptotically minimizes the empirical risk, i.e., to find a sequence of functions { f ^ n } n = 1 ∞ {\displaystyle \{{\hat {f}}_{n}\}_{n=1}^{\infty }} that satisfies lim n → ∞ P ( I P ( f ^ n ) − inf f ∈ F I P ( f ) > ϵ ) = 0 {\displaystyle \lim _{n\rightarrow \infty }\mathbb {P} (I_{P}({\hat {f}}_{n})-\inf _{f\in {\mathcal {F}}}I_{P}(f)>\epsilon )=0} One usual algorithm to find such a sequence is through empirical risk minimization. === Learnable function class === We can make the condition given in the above equation stronger by requiring that the convergence is uniform for all probability distributions. That is: The intuition behind the more strict requirement is as such: the rate at which sequence { f ^ n } {\displaystyle \{{\hat {f}}_{n}\}} converges to the minimizer of the expected risk can be very different for different P ( x , y ) {\displaystyle P(x,y)} . Because in real world the true distribution P {\displaystyle P} is always unknown, we would want to select a sequence that performs well under all cases. However, by the no free lunch theorem, such a sequence that satisfies (1) does not exist if F {\displaystyle {\mathcal {F}}} is too complex. This means we need to be careful and not allow too "many" functions in F {\displaystyle {\mathcal {F}}} if we want (1) to be a meaningful requirement. Specifically, function classes that ensure the existence of a sequence { f ^ n } {\displaystyle \{{\hat {f}}_{n}\}} that satisfies (1) are known as learnable classes. It is worth noting that at least for supervised classification and regression problems, if a function class is learnable, then the empirical risk minimization automatically satisfies (1). Thus in these settings not only do we know that the problem posed by (1) is solvable, we also immediately have an algorithm that gives the solution. == Interpretations == If the true relationship between y {\displaystyle y} and x {\displaystyle x} is y ∼ f ∗ ( x ) {\displaystyle y\sim f^{}(x)} , then by selecting the appropriate loss function, f ∗ {\displaystyle f^{}} can always be expressed as the minimizer of the expected loss across all possible functions. That is, f ∗ = arg min f ∈ F ∗ I P ( f ) {\displaystyle f^{}=\arg \min _{f\in {\mathcal {F}}^{}}I_{P}(f)} Here we let F ∗ {\displaystyle {\mathcal {F}}^{}} be the collection of all possible functions mapping X {\displaystyle {\mathcal {X}}} onto Y {\displaystyle {\mathcal {Y}}} . f ∗ {\displaystyle f^{}} can be interpreted as the actual data generating mechanism. However, the no free lunch theorem tells us that in practice, with finite samples we cannot hope to search for the expected risk minimizer over F ∗ {\displaystyle {\mathcal {F}}^{}} . Thus we often consider a subset of F ∗ {\displaystyle {\mathcal {F}}^{}} , F {\displaystyle {\mathcal {F}}} , to carry out searches on. By doing so, we risk that f ∗ {\displaystyle f^{}} might not be an element of F {\displaystyle {\mathcal {F}}} . This tradeoff can be mathematically expressed as In the above decomposition, part ( b ) {\displaystyle (b)} does not depend on the data and is non-stochastic. It describes how far away our assumptions ( F {\displaystyle {\mathcal {F}}} ) are from the truth ( F ∗ {\displaystyle {\mathcal {F}}^{}} ). ( b ) {\displaystyle (b)} will be strictly greater than 0 if we make assumptions that are too strong ( F {\displaystyle {\mathcal {F}}} too small). On the other hand, failing to put enough restrictions on F {\displaystyle {\mathcal {F}}} will cause it to be not learnable, and part ( a ) {\displaystyle (a)} will not stochastically converge to 0. This is the well-known overfitting problem in statistics and machine learning literature. == Example: Tikhonov regularization == A good example where learnable classes are used is the so-called Tikhonov regularization in reproducing kernel Hilbert space (RKHS). Specifically, let F ∗ {\displaystyle {\mathcal {F^{}}}} be an RKHS, and | | ⋅ | | 2 {\displaystyle ||\cdot ||_{2}} be the norm on F ∗ {\displaystyle {\mathcal {F^{}}}} given by its inner product. It is shown in that F = { f : | | f | | 2 ≤ γ } {\displaystyle {\mathcal {F}}=\{f:||f||_{2}\leq \gamma \}} is a learnable class for any finite, positive γ {\displaystyle \gamma } . The empirical minimization algorithm to the dual form of this problem is arg min f ∈ F ∗ { ∑ i = 1 n L ( f ( x i ) , y i ) + λ | | f | | 2 } {\displaystyle \arg \min _{f\in {\mathcal {F}}^{}}\left\{\sum _{i=1}^{n}L(f(x_{i}),y_{i})+\lambda ||f||_{2}\right\}} This was first introduced by Tikhonov to solve ill-posed problems. Many statistical learning algorithms can be expressed in such a form (for example, the well-known ridge regression). The tradeoff between ( a ) {\displaystyle (a)} and ( b ) {\displaystyle (b)} in (2) is geometrically more intuitive with Tikhonov regularization in RKHS. We can consider a sequence of { F γ } {\displaystyle \{{\mathcal {F}}_{\gamma }\}} , which are essentially balls in F ∗ {\displaystyle {\mathcal {F^{}}}} with centers at 0. As γ {\displaystyle \gamma } gets larger, F γ {\displaystyle {\mathcal {F}}_{\gamma }} gets closer to the entire space, and ( b ) {\displaystyle (b)} is likely to become smaller. However we will also suffer smaller convergence rates in ( a ) {\displaystyle (a)} . The way to choose an optimal γ {\displaystyle \gamma } in finite sample settings is usually through cross-validation. == Relationship to empirical process theory == Part ( a ) {\displaystyle (a)} in (2) is closely linked to empirical process theory in statistics, where the empirical risk { ∑ i = 1 n L ( y i , f ( x i ) ) , f ∈ F } {\displaystyle \{\sum _{i=1}^{n}L(y_{i},f(x_{i})),f\in {\mathcal {F}}\}} are known as empirical processes. In this field, the function class F {\displaystyle {\mathcal {F}}} that satisfies the stochastic convergence are known as uniform Glivenko–Cantelli classes. It has been shown that under certain regularity conditions, learnable classes and uniformly Glivenko-Cantelli classes are equivalent. Interplay between ( a ) {\displaystyle (a)} and ( b ) {\displaystyle (b)} in statistics literature is often known as the bias-variance tradeoff. However, note that in the authors gave an example of stochastic convex optimization for General Setting of Learning where learnability is not equivalent with uniform convergence.
Yaron Singer
Yaron Singer is a computer scientist and entrepreneur whose work has focused on algorithms, machine learning, optimization, and artificial intelligence security. He was the Gordon McKay Professor of Computer Science and Applied Mathematics at Harvard University and co-founded Robust Intelligence, an artificial intelligence security company acquired by Cisco Systems in 2024. == Education == Singer received a PhD in computer science from the University of California, Berkeley under the supervision of Christos Papadimitriou. == Academic career == Singer was a postdoctoral research scientist at Google Research. Singer joined the computer science faculty at Harvard John A. Paulson School of Engineering and Applied Sciences in 2013 and became a full professor in 2019. == Research == Singer's research has focused on algorithms and machine learning, including optimization, algorithmic mechanism design, and adversarial machine learning. His doctoral work studied computational limits in algorithmic mechanism design, including truthful mechanisms and budget-feasible mechanisms. In optimization, Singer co-authored work on submodular optimization and parallel algorithms for large-scale data processing. Singer has also worked on adversarial machine learning, including attacks that use small perturbations or noise to affect the behavior of machine learning systems. == Entrepreneurship == In 2020, Singer co-founded Robust Intelligence Kojin Oshiba. Harvard SEAS reported that the company raised $14 million that year, and TechCrunch reported in 2021 that the company raised a $30 million Series B round led by Tiger Global. The company developed tools for testing AI models and detecting failures before or during deployment. TechCrunch described its RIME product as using an "AI firewall" to stress-test models. In 2024, Cisco Systems acquired Robust Intelligence. CTech reported that Cisco had not disclosed the purchase amount when the acquisition was announced, and later reported the deal value as $400 million. In 2025, Cisco launched Foundation AI, a Cisco team focused on AI for cybersecurity. Techzine reported that Singer led the team and was Cisco's VP of AI and Security. == Recognition == Singer has received a Sloan Research Fellowship, an NSF CAREER Award, a Google Faculty Research Award, and a Facebook Faculty Award. As a graduate student, he received Microsoft Research and Facebook fellowships. In 2012, he received the Best Student Paper Award at the ACM International Conference on Web Search and Data Mining for "How to Win Friends and Influence People, Truthfully: Influence Maximization Mechanisms for Social Networks."
Korpusomat
Korpusomat - a tool for creating and searching electronic language corpora, created at the Institute of Computer Science of the Polish Academy of Sciences. Korpusomat is a fourth generation corpus tool. It is a web application, which eliminates the need to store data sets on the user's own computer. The corpus is created either by adding text files from the local drive (in any language and format), or by indicating websites from which texts are to be downloaded. Then, the corpus is annotated automatically on several levels: morphosyntantic, named entities recognition (e.g. geographical names or people) and partial syntantic information (which also allows for the visualization of dependency trees). The finished corpus can be edited, shared with other users, and searched. There are also a number of functions offering statistical summaries of the collected texts
The Best Free AI Blog Writer for Beginners
Looking for the best AI blog writer? An AI blog writer is software that uses machine learning to help you get more done — it can save you hours every week by automating repetitive work. Most options offer a generous free tier, with paid plans unlocking higher limits, faster processing, and team features. Whether you are a beginner or a pro, the right AI blog writer slots into your workflow and pays for itself fast. Read on for hands-on impressions, pricing tiers, and the standout features that matter.
Actor-critic algorithm
The actor-critic algorithm (AC) is a family of reinforcement learning (RL) algorithms that combine policy-based RL algorithms such as policy gradient methods, and value-based RL algorithms such as value iteration, Q-learning, SARSA, and TD learning. An AC algorithm consists of two main components: an "actor" that determines which actions to take according to a policy function, and a "critic" that evaluates those actions according to a value function. Some AC algorithms are on-policy, some are off-policy. Some apply to either continuous or discrete action spaces. Some work in both cases. == Overview == The actor-critic methods can be understood as an improvement over pure policy gradient methods like REINFORCE via introducing a baseline. === Actor === The actor uses a policy function π ( a | s ) {\displaystyle \pi (a|s)} , while the critic estimates either the value function V ( s ) {\displaystyle V(s)} , the action-value Q-function Q ( s , a ) , {\displaystyle Q(s,a),} the advantage function A ( s , a ) {\displaystyle A(s,a)} , or any combination thereof. The actor is a parameterized function π θ {\displaystyle \pi _{\theta }} , where θ {\displaystyle \theta } are the parameters of the actor. The actor takes as argument the state of the environment s {\displaystyle s} and produces a probability distribution π θ ( ⋅ | s ) {\displaystyle \pi _{\theta }(\cdot |s)} . If the action space is discrete, then ∑ a π θ ( a | s ) = 1 {\displaystyle \sum _{a}\pi _{\theta }(a|s)=1} . If the action space is continuous, then ∫ a π θ ( a | s ) d a = 1 {\displaystyle \int _{a}\pi _{\theta }(a|s)da=1} . The goal of policy optimization is to improve the actor. That is, to find some θ {\displaystyle \theta } that maximizes the expected episodic reward J ( θ ) {\displaystyle J(\theta )} : J ( θ ) = E π θ [ ∑ t = 0 T γ t r t ] {\displaystyle J(\theta )=\mathbb {E} _{\pi _{\theta }}\left[\sum _{t=0}^{T}\gamma ^{t}r_{t}\right]} where γ {\displaystyle \gamma } is the discount factor, r t {\displaystyle r_{t}} is the reward at step t {\displaystyle t} , and T {\displaystyle T} is the time-horizon (which can be infinite). The goal of policy gradient method is to optimize J ( θ ) {\displaystyle J(\theta )} by gradient ascent on the policy gradient ∇ J ( θ ) {\displaystyle \nabla J(\theta )} . As detailed on the policy gradient method page, there are many unbiased estimators of the policy gradient: ∇ θ J ( θ ) = E π θ [ ∑ 0 ≤ j ≤ T ∇ θ ln π θ ( A j | S j ) ⋅ Ψ j | S 0 = s 0 ] {\displaystyle \nabla _{\theta }J(\theta )=\mathbb {E} _{\pi _{\theta }}\left[\sum _{0\leq j\leq T}\nabla _{\theta }\ln \pi _{\theta }(A_{j}|S_{j})\cdot \Psi _{j}{\Big |}S_{0}=s_{0}\right]} where Ψ j {\textstyle \Psi _{j}} is a linear sum of the following: ∑ 0 ≤ i ≤ T ( γ i R i ) {\textstyle \sum _{0\leq i\leq T}(\gamma ^{i}R_{i})} . γ j ∑ j ≤ i ≤ T ( γ i − j R i ) {\textstyle \gamma ^{j}\sum _{j\leq i\leq T}(\gamma ^{i-j}R_{i})} : the REINFORCE algorithm. γ j ∑ j ≤ i ≤ T ( γ i − j R i ) − b ( S j ) {\textstyle \gamma ^{j}\sum _{j\leq i\leq T}(\gamma ^{i-j}R_{i})-b(S_{j})} : the REINFORCE with baseline algorithm. Here b {\displaystyle b} is an arbitrary function. γ j ( R j + γ V π θ ( S j + 1 ) − V π θ ( S j ) ) {\textstyle \gamma ^{j}\left(R_{j}+\gamma V^{\pi _{\theta }}(S_{j+1})-V^{\pi _{\theta }}(S_{j})\right)} : TD(1) learning. γ j Q π θ ( S j , A j ) {\textstyle \gamma ^{j}Q^{\pi _{\theta }}(S_{j},A_{j})} . γ j A π θ ( S j , A j ) {\textstyle \gamma ^{j}A^{\pi _{\theta }}(S_{j},A_{j})} : Advantage Actor-Critic (A2C). γ j ( R j + γ R j + 1 + γ 2 V π θ ( S j + 2 ) − V π θ ( S j ) ) {\textstyle \gamma ^{j}\left(R_{j}+\gamma R_{j+1}+\gamma ^{2}V^{\pi _{\theta }}(S_{j+2})-V^{\pi _{\theta }}(S_{j})\right)} : TD(2) learning. γ j ( ∑ k = 0 n − 1 γ k R j + k + γ n V π θ ( S j + n ) − V π θ ( S j ) ) {\textstyle \gamma ^{j}\left(\sum _{k=0}^{n-1}\gamma ^{k}R_{j+k}+\gamma ^{n}V^{\pi _{\theta }}(S_{j+n})-V^{\pi _{\theta }}(S_{j})\right)} : TD(n) learning. γ j ∑ n = 1 ∞ λ n − 1 1 − λ ⋅ ( ∑ k = 0 n − 1 γ k R j + k + γ n V π θ ( S j + n ) − V π θ ( S j ) ) {\textstyle \gamma ^{j}\sum _{n=1}^{\infty }{\frac {\lambda ^{n-1}}{1-\lambda }}\cdot \left(\sum _{k=0}^{n-1}\gamma ^{k}R_{j+k}+\gamma ^{n}V^{\pi _{\theta }}(S_{j+n})-V^{\pi _{\theta }}(S_{j})\right)} : TD(λ) learning, also known as GAE (generalized advantage estimate). This is obtained by an exponentially decaying sum of the TD(n) learning terms. === Critic === In the unbiased estimators given above, certain functions such as V π θ , Q π θ , A π θ {\displaystyle V^{\pi _{\theta }},Q^{\pi _{\theta }},A^{\pi _{\theta }}} appear. These are approximated by the critic. Since these functions all depend on the actor, the critic must learn alongside the actor. The critic is learned by value-based RL algorithms. For example, if the critic is estimating the state-value function V π θ ( s ) {\displaystyle V^{\pi _{\theta }}(s)} , then it can be learned by any value function approximation method. Let the critic be a function approximator V ϕ ( s ) {\displaystyle V_{\phi }(s)} with parameters ϕ {\displaystyle \phi } . The simplest example is TD(1) learning, which trains the critic to minimize the TD(1) error: δ i = R i + γ V ϕ ( S i + 1 ) − V ϕ ( S i ) {\displaystyle \delta _{i}=R_{i}+\gamma V_{\phi }(S_{i+1})-V_{\phi }(S_{i})} The critic parameters are updated by gradient descent on the squared TD error: ϕ ← ϕ − α ∇ ϕ ( δ i ) 2 = ϕ + α δ i ∇ ϕ V ϕ ( S i ) {\displaystyle \phi \leftarrow \phi -\alpha \nabla _{\phi }(\delta _{i})^{2}=\phi +\alpha \delta _{i}\nabla _{\phi }V_{\phi }(S_{i})} where α {\displaystyle \alpha } is the learning rate. Note that the gradient is taken with respect to the ϕ {\displaystyle \phi } in V ϕ ( S i ) {\displaystyle V_{\phi }(S_{i})} only, since the ϕ {\displaystyle \phi } in γ V ϕ ( S i + 1 ) {\displaystyle \gamma V_{\phi }(S_{i+1})} constitutes a moving target, and the gradient is not taken with respect to that. This is a common source of error in implementations that use automatic differentiation, and requires "stopping the gradient" at that point. Similarly, if the critic is estimating the action-value function Q π θ {\displaystyle Q^{\pi _{\theta }}} , then it can be learned by Q-learning or SARSA. In SARSA, the critic maintains an estimate of the Q-function, parameterized by ϕ {\displaystyle \phi } , denoted as Q ϕ ( s , a ) {\displaystyle Q_{\phi }(s,a)} . The temporal difference error is then calculated as δ i = R i + γ Q θ ( S i + 1 , A i + 1 ) − Q θ ( S i , A i ) {\displaystyle \delta _{i}=R_{i}+\gamma Q_{\theta }(S_{i+1},A_{i+1})-Q_{\theta }(S_{i},A_{i})} . The critic is then updated by θ ← θ + α δ i ∇ θ Q θ ( S i , A i ) {\displaystyle \theta \leftarrow \theta +\alpha \delta _{i}\nabla _{\theta }Q_{\theta }(S_{i},A_{i})} The advantage critic can be trained by training both a Q-function Q ϕ ( s , a ) {\displaystyle Q_{\phi }(s,a)} and a state-value function V ϕ ( s ) {\displaystyle V_{\phi }(s)} , then let A ϕ ( s , a ) = Q ϕ ( s , a ) − V ϕ ( s ) {\displaystyle A_{\phi }(s,a)=Q_{\phi }(s,a)-V_{\phi }(s)} . Although, it is more common to train just a state-value function V ϕ ( s ) {\displaystyle V_{\phi }(s)} , then estimate the advantage by A ϕ ( S i , A i ) ≈ ∑ j ∈ 0 : n − 1 γ j R i + j + γ n V ϕ ( S i + n ) − V ϕ ( S i ) {\displaystyle A_{\phi }(S_{i},A_{i})\approx \sum _{j\in 0:n-1}\gamma ^{j}R_{i+j}+\gamma ^{n}V_{\phi }(S_{i+n})-V_{\phi }(S_{i})} Here, n {\displaystyle n} is a positive integer. The higher n {\displaystyle n} is, the more lower is the bias in the advantage estimation, but at the price of higher variance. The Generalized Advantage Estimation (GAE) introduces a hyperparameter λ {\displaystyle \lambda } that smoothly interpolates between Monte Carlo returns ( λ = 1 {\displaystyle \lambda =1} , high variance, no bias) and 1-step TD learning ( λ = 0 {\displaystyle \lambda =0} , low variance, high bias). This hyperparameter can be adjusted to pick the optimal bias-variance trade-off in advantage estimation. It uses an exponentially decaying average of n-step returns with λ {\displaystyle \lambda } being the decay strength. == Variants == Asynchronous Advantage Actor-Critic (A3C): Parallel and asynchronous version of A2C. Soft Actor-Critic (SAC): Incorporates entropy maximization for improved exploration. Deep Deterministic Policy Gradient (DDPG): Specialized for continuous action spaces.
Jollo
Jollo was an online machine translation service where users could instantly translate texts into 23 languages, request human translations from a community of volunteers around the world, and compare the correctness of several leading machine translation websites. It was discontinued in 2012. == System == Jollo was a free Web 2.0 website that attempted to improve the way in which people translate online through the use of existing machine translation websites and a community of volunteers who correct and rate translations. The system relied on a similar methodology as computer-assisted translation to ensure translation quality, and featured a public translation memory that records past translations. Jollo received some notable media attention, including in The Daily Telegraph. According to the blog KillerStartups, Jollo combined the benefits of the speed of machine translations and human reviews to ensure translation quality. According to Jeffrey Hill from The English Blog, the community features made Jollo an interesting alternative to other online translation services. == Development == The Jollo website was classified as beta. It was developed using LAMP and was praised for its colorful graphics and simple user interface. Jollo offered a simple web-based API that could be used for translations. For example, the URL: http://www.jollo.com/translate.php?st=I%20love%20you&sl=en&tl=zh was used to translate the sentence "I love you" from English into Chinese.