# Tag Info

7

Yes this is done routinely. For example this is how the YOLO object detection and classifier system works, to give a real-world for example. In YOLO, the "non-object" classification is "background" i.e. any image segment that doesn't contain one of the types of object we are interested in. In general, you can add an "other" class to any classifier, provided ...

4

I suppose the most common part where it will be used is in the initialization of weights before training; the best ways currently known to do that involve randomness. If you use Dropout during training (randomly setting some activation levels to zero to combat overfitting), that also involves randomness, so your seed could also have influence there. Dropout ...

4

I will try to answer all three questions to the best of my ability. Is this basically true? Quick googling just brought me to a lot of papers trying to fit decision trees into incremental learning. The problem with decision trees in an online learning setting is that the model should be able to update when experiences (i.e. state, action, reward, new state)...

3

This seems to fall broadly into the regime of a classification problem as you want to classify an outgoing communication as "contains proprietary information" or "does not contain proprietary information". As such, any classification approach could be applied. Neural Networks certainly seem like a valid approach, but you might also get good mileage out of ...

3

In computational learning theory, a learning algorithm (or learner) $A$ is an algorithm that chooses a hypothesis (which is a function) $h: \mathcal{X} \rightarrow \mathcal{Y}$, where $\mathcal{X}$ is the input space and $\mathcal{Y}$ is the target space, from the hypothesis space $H$. For example, consider the task of image classification (e.g. MNIST). You ...

3

Depends on what does 1 represent in your task. If you are trying to predict household prices and 1 represents \$1, I think the average validation loss is good. If 1 represents \$10000 in this case, probably something is not right. But remember that there are 2 parts contributing to the overall loss. The mse loss and the l2 penalty loss. (Also remember that ...

3

The validation loss settles exactly at an error of one. Probably means there's something off with either the kind of data validation set has or with something in the training. An exact validation loss of one almost definitely means there's something off. I'd recommend before doing anything thoroughly go through your data or see if there's anything to debug ...

2

Adding to SmallChess's answer , Larger trees(with many nodes) are too adapted to the training set, as a small change in the input train data might cause the trees to change very much and hence change the estimate value too much.This is mainly due to the hierarchical structures of trees(because a change in a higher node may cause all lower nodes to change). ...

2

The bigger your tree is the more overfitting your model is. In machine learning, we always prefer a simpler model unless there is good reason to go for complication.

2

For your first question,take a look at using a skip grab model to find what the abbreviated text is. The skip gram model turns a word into a vector which allows it to be processed by other machine learning algorithms. Or , alternatively you can do some really cool addition and subtraction problems with the resulting vectors. With the skip gram model in your ...

2

There is no transformation between epochs. One full iteration over the training set is considered an epoch. Lets assume: We're considering a gradient descent in a space without local optima. This means that if you'd plot the errors we calculate below, this plot has but one lowest point. The perceptron is intended to model a linearely seperable function. So,...

2

I wouldn't focus only on "deep learning" unless you have some specific reason for doing so. There may be other techniques which could be as effective, or more effective. One approach I've seen used for something similar was Inductive Logic Programming. For one example of using ILP to reason about elements of biochemistry, see this paper That's not ...

2

The The Oxford Companion to Chess has entries on only 700 named openings, and lists another 1327 opening variations in the index, and I wouldn't be surprised if someone out there had them all memorized. For an algorithm, however, storing that number of openings is trivial, and Chess algorithms traditionally made use of high-quality "game books" which are ...

2

If you can remember everything and there's no randomisation in your outcome like chess, there is absolutely no reason not to do that. Anybody who can remember all the possible board configurations in chess, by definition plays perfect chess. A perfect player would never lose. Unfortunately, most practical problems can't be solved by brute-force, and that ...

2

Any AI algorithm depends on the environment, and available actuators and sensors. In our case, the environment is a road, street, etc. The primary actuator includes wheels (or legs) of the robot. Sensors include a camera, sonar system, etc. A simple Model-based reflex algorithm can work in your case: function MODEL-BASED-REFLEX-AGENT(percept) returns an ...

2

Don’t think about it as the $w_{(n)}-w_{(n+1)}$ being proportional to something. Think about it this way: I'm now at $w_{(n)}$. Where do I want to be at timestep, so that the error decreases? For that, I need to know how the error changes when I make small steps to the left or right of $w_{(n)}$. If $E$ increases as I increase $w$ (that is, if $\frac{\... 2 It is calculated the same way$b_1$is calculated. Nearly following your notation, say your multiple linear regression function is$H(X_i) = b_0 + b_1x_{1,i} + ...+ b_nx_{n, i}$for data instance$X_i=x_{1,i},...,x_{n, i}$and weights$b_0,...,b_n$. And say your error function is$E(X,Y) = \sum_i(H(X_i)-Y_i)^2$where$X$is the collection of all data ... 2 Welcome to AI.SE @GundamOfOasis! Your intuition is right: this is fundamentally a problem for combinatorial search. You're also right that problems are created by the fact that not every move is valid at state. To fix this, you need to add a function that can determine whether a given state is valid or not, in addition to the usual function that checks ... 2 Yes! If you read ahead to the chapters in reinforcement learning in the same book, you'll see that the wompus world appears again there. Techniques like Q-learning can be used to solve it, and since Q-learning involves learning the shape of a function, a neural network can be employed as a function approximator. The basic idea is to treat this problem as an ... 2 I found someone that has done this thing! You can hear a good explanation in Marcus Hutter's answer to this question about rewards given to AIXI. He describes a work that seems to be referring to this paper: Universal Knowledge-Seeking Agents for Stochastic Environments I'll edit this answer later with a full explanation of the approach, but essentially ... 2 There are two factors that will change the ability of a deep neural network to fit a given dataset: either you need more data, or a deeper and wider network. Since the pattern is only 2-d, it can likely be approximated by some sort of simple periodic function. A DNN can approximate periodic functions pretty well, so the issue is probably that you don't have ... 2$f(x) = x^2 + b$is a polynomial (more precisely, a parabola) so it is continuous, thus, a neural network (with at least one hidden layer) should be able to approximate that function (given the universal approximation theorem). After a very quick look at your code, I noticed you aren't using an activation function for your dense layers (i.e. your activation ... 2 @The Pointer the$2^n$came from the question: How many function do we need to have if each of the$n$inputs can be missing? example:$f_1(\text{missing}, x_2, x_3, \dots, x_n)$for$x_1$missing$f_2(x_1, x_2, \text{missing}, x_4, \text{missing}, \dots, x_n)$for$x_3$and$x_5$missing. So this problem is a combinatorial one and the event for each$x_i... 2 Any algorithm that uses data (in some form) to improve some performance measure (aka objective function), or to find some function, can be considered a machine learning algorithm. See this answer for more complete definitions of ML. k-means does that. It uses the data to find some division of the data itself into groups, in order to maximize some objective ... 1 for this kind of ml training, you will need a ton of data first, at least in the thousands. If you have a bot program that fetches those data for you, AI is the way to go. I'm not sure how else you would do it though. To train the nn you will need the inputs(the post) and the targets(the rating you want it to output). The targets could be anything you want, ... 1 You are basically describing the way Google Translate works. There has been a lot of research in text alignment in the area of multi-lingual corpus linguistics. An early paper (with sourcode) is Gale and Church's A Program for Aligning Sentences in Bilingual Corpora (PDF). In linguistics these are called parallel texts. On the wikipedia page you will find ... 1 Here is one idea. I'll start with a more specific "mathematical singularity", defined as an algorithm that can do the following in N hours or less (for allN >= 1\$): State equivalent versions (up to notional differences) of all mathematical theorems/conjectures that humans will read and understand in N*20 years after 2018 that can be stated formally in ...

1

If you don't have non-examples of your pattern and don't have some kind of heuristic guide, unfortunately the answer is that you can't. "All sentences" will always be 100% compatible with your examples, and you'll never be able to collect evidence that disconfirms (or even decreases the likelihood) of that hypothesis. Even if you rule out that hypothesis by ...

1

What you have is called a classification problem with categorical features. That is, the features can be represented numerically, but the numbers have no relative meaning. Algorithms that rely on smooth function approximation will probably not work well here. These would include classic approaches to regression, and also function approximation via a neural ...

1

The point of pseudoRNG is to be unmodable and unpredictable, making it hard to train an AI to learn. It would more likely be useful and more efficient to have the equation that the game uses for generation available, so that you can manually make the check, or to just have a list of the loop if the pseudoRNG is based on the time elapsed.

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