Activa JavaScript para disfrutar de los vídeos de la Mediateca.
Curso IA programamos.es -- cómo entienden el texto los ordenadores_eng - Contenido educativo
Ajuste de pantallaEl ajuste de pantalla se aprecia al ver el vídeo en pantalla completa. Elige la presentación que más te guste:
As we already know, computers can only work with numbers, and yet, there are computer systems that are capable of understanding our texts.
00:00:00
How does this happen?
00:00:08
What is used is a mechanism to translate words or phrases into a numerical representation known as embeddings.
00:00:10
As Romea Jeremy Howard says in his book "AI Applications Without a PhD",
00:00:17
the artificial intelligence community sometimes likes to use rather pompous names for concepts that are actually very simple.
00:00:22
And with embeddings, this is somewhat the case. Let's see how they are built.
00:00:30
Imagine we are in a situation where a numerical representation has already been assigned to a set of words, using two numbers.
00:00:34
Where would we place the word "apple"?
00:00:42
Near position A there are several round objects. Near position B there are words related to constructions.
00:00:45
But at position C, we would have the word "apple" close to others related to fruits.
00:00:53
This would be a good location, since the goal of embeddings is for similar words
00:00:58
to correspond to nearby points, and words that are different to correspond
00:01:02
to distant points.
00:01:07
Let's see another example.
00:01:09
Suppose we have already assigned a numerical representation to the words "dog", "puppy", and
00:01:11
"calf".
00:01:17
Where would we place the word "cow"?
00:01:18
All three positions could make sense, but if we place it at position
00:01:20
C, we would be capturing some relationships between the words, which is precisely another
00:01:25
goal of embeddings. In this case, we would be capturing two analogies.
00:01:30
On one hand, "puppy" is to "dog" as "calf" is to "cow". And on the other, "puppy" is to "calf",
00:01:36
as "dog" is to "cow". Thus, this embedding would be capturing two properties of the
00:01:42
words, age and size. And basically, these are embeddings. What happens is that the
00:01:49
ones we use in real applications have hundreds or thousands of dimensions, meaning
00:01:54
that a word is translated into a vector of hundreds or thousands of numbers.
00:01:58
As detailed in the article associated with this video, these embeddings allow
00:02:04
performing visualizations and classroom activities that are very interesting and that
00:02:08
could be the 21st-century equivalent of learning to explore a dictionary.
00:02:13
But these word embeddings have certain limitations when it comes to
00:02:19
recognizing phrases, since the same word can mean different things depending on the
00:02:22
context. Fortunately, since transformers were born with their attention mechanism that
00:02:28
allows understanding the context, we also have embeddings that are capable of assigning a
00:02:34
numerical representation to complete phrases coherently. Thus, we can see that the phrase
00:02:39
"I like basketball more than anything" is semantically closer to "I love basketball"
00:02:45
than the phrase "I love football", even though these last two share more words
00:02:51
in common. And there are even multilingual phrase embeddings where phrases that
00:02:56
mean the same thing in different languages receive a close numerical representation.
00:03:02
As we will see in future installments, these word and phrase embeddings are the basis
00:03:07
of large language models like GPT-3 and Bloom. But until we get there, don't
00:03:12
stop playing with the challenges and tasks we propose on our website, as they
00:03:17
will allow you to interact directly with the internal workings of many
00:03:22
of the artificial intelligence systems we use daily.
00:03:26
- Idioma/s:
- Idioma/s subtítulos:
- Autor/es:
- Programamos.es
- Subido por:
- David G.
- Licencia:
- Reconocimiento
- Visualizaciones:
- 18
- Fecha:
- 29 de marzo de 2024 - 23:08
- Visibilidad:
- Público
- Centro:
- IES MARIE CURIE Loeches
- Duración:
- 03′ 30″
- Relación de aspecto:
- 1.78:1
- Resolución:
- 854x480 píxeles
- Tamaño:
- 17.73 MBytes
