28

Here are three survey papers that examine the use of machine learning in time series forecasting: "An Empirical Comparison of Machine Learning Models for Time Series Forecasting" by Ahmed, Atiya, El Gayar, and El-shishiny provides an empirical comparison of several machine learning algorithms, including: "...multilayer perceptron, Bayesian neural ...


18

Neural Networks are not magic. If you treat them like they are and just throw data at them without thinking you're going to have a very bad time. You need to stop and ask youself "Is milliseconds since 1970 actually going to be predictive of the event I'm interested in?" The answer you should arrive at immediately is no. Why? For every instance you ...


18

There is no "official Turing test" so there's no concept of "officially pass[ing] the test". Turing described a methodology that one might use to evaluate artificial intelligences. The organizers of the event that Eugene Goostman won implemented that methodology in a particular way and the program satisfied the criteria the organizers had chosen. In that ...


18

The Church-Turing thesis says that the informal notion of an algorithm as a sequence of instructions coincides with Turing machines. Equivalently, it says that any reasonable model of computation has the same power as Turing machines. An artificial intelligence is a computer program, i.e., an algorithm. If the Church-Turing thesis holds, then you could ...


17

The halting problem is not a statement about intelligence (human or artificial) it is a statement about the limits of mathematics. It is an historically important example of an undecidable problem. An artificial (or human) intelligence can certainly look for and find many sorts of different infinite loops in real programs. And the halting problem doesn't ...


17

Recoverable, as we can pour all water back to the 12L jug to restore the original state, hence any state derived thereof (by following the same steps from the start). The problem is, also, solvable: [12, 0, 0] [4, 8, 0] [4, 3, 5] [9, 3, 0] [9, 0, 3] [1, 8, 3] [1, 6, 5] [6, 6, 0] # <-- SOLVED; 7 steps DETAILS: The problem stated as-is is ...


16

If the heuristic function is not admissible, than we can have an estimation that is bigger than the actual path cost from some node to a goal node. If this higher path cost estimation is on the least cost path (that we are searching for), the algorithm will not explore it and it may find another (not least cost) path to the goal. Look at this simple example....


16

The whole state space for chess is enormous - it can be roughly estimated as 1043 (Shannon number (Shannon, 1950), (Wikipedia)). The idea you present - Reinforcement Learning agents playing with each other to learn the game - was successfully applied to Backgammon - TD-Gammon (Tesauro, 1995), (Chapter in Reinforcement Learning by Sutton&Barto). It also ...


12

This is an expansion of this presentation. Because the state graph consists of two disconnected components of equal size. Without loss of generality we can assume that the target state is $1\;2\;3\;...\;15\;\Box$. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 * Given a state $S$ a permutation inversion is a tile $T_i$ that is placed after $T_j$ but $i &...


10

For the field of A.I. and machine learning, I would recommend you to explore and learn more about these topics: Statistics Probability Stochastic processes Bayesian Data Analysis Convex Optimization Graph Theory With your math background, you could easily pick any good machine learning book and learn the required math that you don't have as you go. Kevin ...


10

Although Alan Turing is of course a very important computer scientist, the Turing test is only superficially related to computer science. It is more related to philosophy. As long as machines exist people have wondered whether it is possible to construct a machine that can think. But to answer this question we must first answer the question what it means -- ...


9

As noted by Thomas Klimpel in the comments, a certain acceptance probability is often used, which is equal to say $0.8$. The following is a simple iterative method to find a suitable initial temperature, proposed by Ben-Ameur in 2004 [1]. In the following, $t$ is a strictly positive transition, $\max_t$ and $\min_t$ are the states after and before the ...


9

The most common/obvious way is a challenge-response test that is easy for humans but hard for computers (of course, but not only, CAPTCHA). This kind of test is very effective{1} but falls under the HIP (Human Interactive Proofs) area: it's not transparent. Typical, "simple" approaches to distinguish human website traffic from Bot are: time it takes to ...


9

Roughly speaking, over-fitting typically occurs when the ratio $\frac{\text{complexity of the model}}{\text{training set size}} $ is too high. Think of over-fitting as a situation where your model learn the training data by heart instead of learning the big pictures which prevent it from being able to generalized to the test data: this happens when the model ...


9

I think the prizes you're referring to are the Loebner Prize. According to the Wikipedia page (see prior link), the winner for 2014 is 'Rose' by Bruce Wilcox. That program did not win one of the one-time-only prizes, but did get $4,000 in prize money. 'Eugene Goostman' competed in 2005 and 2008, finishing second both times. The competition 'Eugene Goostman'...


8

Perhaps the simplest way to understand the proof is by the idea of a conserved quantity : find some quantity that can be derived from a configuration and show that every move preserves that quantity. A static version of the idea is found in the following old puzzle: Remove the northeast and southwest corner squares from a standard 8x8 checkerboard. Can ...


8

With neural networks, you always need to randomly initialize your weights to break symmetry. If you don't use a non-linear activation function in the hidden units, then you might as well have stayed with a single layer. Your network is now just a composition of two linear functions, which is of course just another linear function. That learning rate seems ...


8

While Anton's answer is absolutely perfect let me try to provide an alternative answer: being admissible means that the heuristic does not overestimate the effort to reach the goal, i.e., $h(n) \leq h^*(n)$ for all $n$ in the state space (in the 8-puzzle, this means just for any permutation of the tiles and the goal you are currently considering) where $h^*(...


8

Connect Four was solved in 1988. The first solution was given by Allen and, in the same year, Allis coded VICTOR which actually won the computer-game olympiad in the category of connect four. I would suggest you to go to Victor Allis' PhD who graduated in September 1994. You can get a copy of his PhD here. In Section 6.3.2 Connect-Four (page 163) you can ...


8

Turing machines are a model of computation, one way of formally defining the concept of an algorithm. While Turing machines are usually defined using barebones input/output capabilities, it is not difficult to extend them so that they can apply to the kind of problems that AI is after. The relation between AI and Turing machines is thus: Artificial ...


7

Back to the definitions: $$H(L\mid A) = \sum_a p(A=a) H(L \mid A=a).$$ As you compute, $P(A=true)=6/8$ and $P(A=false)=2/8$. However, you don't compute $H(L\mid A=true)$ but instead compute $P(L=positive\mid A=true)$. [and the same for $A=false$.]. With standard definition of $H()$ we get, $$H(L\mid A=true) = - 4/6\log_2(4/6) - 2/6\log_2(2/6) = 0....


7

At a stretch it is an expert system (such as fuzzy logic). As you are not running an algorithm to perform feedback onto the decision parameters based on the output, it's not really learning. However, performing feedback is not the only indicator whether an alogirthm is AI. One could argue that if it acts in a way that appears intelligent, that's all that ...


7

The deterministic requirement isn't all that constraining. That just implies your vehicle is certain of the state it's in. That being said, you'll probably want to plan paths in a way that allows you to avoid obstacles. The best way I've seen this done is with sampling-based planners. Steven LaValle wrote the central academic resource on this topic: Planning ...


7

The idea of DeepDream is this: pick some layer from the network (usually a convolutional layer), pass the starting image through the network to extract features at the chosen layer, set the gradient at that layer equal to the activations themselves, and then backpropagate to the image. Why does it make sense? Intuitively, it amplifies the features that are ...


7

Absolutely yes! your arguments are correct. And, as matter of fact, it is very easy to come up with a graph where Bidirectional Dijkstra would expand more nodes than Unidirectional Dijkstra, following your reasoning. Consider a graph $G(V, E)$ where $V$ is the set of vertices, $E$ is the set of edges and you want to find the shortest path between two ...


7

You're understanding of "solve all instances of the Halting Problem" is flawed. All instances means ALL instances. Every single one, ever. For example, no human knows if a Turing Machine verifying the Collatz Conjecture will halt. Bam, right there, your premise that humans can solve "all" incarnations of the Halting problem is flawed. There is no reason to ...


7

I couldn't in a quick Google through the textbooks tell whether recoverability requires immediate, single-move recoverability (ctrl-Z-style) with each undo step being O(1) complexity, and which can be implementable as a sequence of moves in a push/pop stack. They always use guaranteed-single-move-undo examples (8-puzzle, Towers of Hanoi) for their ...


6

Artificial Intelligence is a catch-all term for a large collection of everything from philosophical questions down to very specific programming techniques. At heart, AI comes from the question is it possible to build a machine that displays intelligent behavior to the same level as human beings? This is often clarified with examples like, can we build a ...


6

AI is 99% statistics these days. Learn about probability, and how it intersects with graph theory (bayes nets, etc.). As for cryptography, if you've got number theory, the only real thing I can think of to extend this is group/field theory. In particular, learn about eliptic curves, but I doubt you'd find a math class that taught that that wasn't ...


6

The basic intuition behind initializing weight layers into small (and different) values is just so that the bias of the system is broken and weight values can move along and away and apart to different values. More concretely, you'ld probably want your initial weights to be distinct and have "a small gap" between them, this 'gap' expands out as you go along ...


Only top voted, non community-wiki answers of a minimum length are eligible