Skip to first unread message
Guna Prasaad
Oct 5, 2013, 12:28:39 PM10/5/13
to wncc...@googlegroups.com
Last week Google announced a brand new algorithm for its search engine, called Humming Bird. Although Google has not given away any details, it has said that the new algorithm focuses on ranking information based on a better understanding of search queries.
The search results we currently receive reflect the matching combination of key words that a search phrase contains, rather than the true meaning of the sentence itself. Search results produced by Hummingbird will reflect the full semantic meaning of longer search phrases, and should in theory produce more accurate results. For example Hummingbird will more greatly consider question words like “how” “why”, “where” and “when” in search phrases, in addition to content keywords. Hence Hummingbird moves the emphasis of search from “results” to “answers”. The semantic search capabilities of Hummingbird aim to address this need.
This is something that I could get by random web scraping from blogs on humming bird algorithm. I think this is an interesting topic and many would like to know more about this, especially people who are enrolled in CS635 (Information Retrieval & Mining for Hypertext & the Web). Lets have an informal meetup on this topic, where we can have a discussion on this, based on whatever information each one could gather. Just a small show of hands here.
Nivvedan (நிவேதன்)
Oct 5, 2013, 3:49:42 PM10/5/13
to WnCC Google Group
+1
This email was sent from a device/client/interface that unfortunately cannot sign emails.
Trust the authenticity of this message at your own peril. If doubtful/paranoid, request me for a signed email and I shall get back to you as soon as I can.
2013/10/5 Guna Prasaad <guna...@gmail.com>
Last week Google announced a brand new algorithm for its search engine, called Humming Bird. Although Google has not given away any details, it has said that the new algorithm focuses on ranking information based on a better understanding of search queries.The search results we currently receive reflect the matching combination of key words that a search phrase contains, rather than the true meaning of the sentence itself. Search results produced by Hummingbird will reflect the full semantic meaning of longer search phrases, and should in theory produce more accurate results. For example Hummingbird will more greatly consider question words like “how” “why”, “where” and “when” in search phrases, in addition to content keywords. Hence Hummingbird moves the emphasis of search from “results” to “answers”. The semantic search capabilities of Hummingbird aim to address this need.This is something that I could get by random web scraping from blogs on humming bird algorithm. I think this is an interesting topic and many would like to know more about this, especially people who are enrolled in CS635 (Information Retrieval & Mining for Hypertext & the Web). Lets have an informal meetup on this topic, where we can have a discussion on this, based on whatever information each one could gather. Just a small show of hands here.
--
--
The website for the club is http://stab-iitb.org/wncc
To post to this group, send email to wncc...@googlegroups.com
---
You received this message because you are subscribed to the Google Groups "Web and Coding Club IIT Bombay" group.
To unsubscribe from this group and stop receiving emails from it, send an email to wncc_iitb+...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.
Dilawar Singh
Oct 6, 2013, 12:33:18 AM10/6/13
to wncc...@googlegroups.com
It may well turned out to be a collection of algorithms rather than "a new
algorithm", though technically one can call a collection of algorithms "an
algorithm".
What happens during search these days? I am guessing; do a search the way you
have been doing so far, then find a pattern in corpus using, let say simulating
annealing; check if the result make sense, if not do a machine learning and
build a database; etc. etc..
And when product is ready, give it a nice name. This trick is quite popular
these days. Intel announced "3D transistors" as a new keyword. Now which
transistor is not a 3D transistor.
A few days ago, there was a talk here on a "new" approach to remove extra
terrestrial pathogens from human body; the approach was titled "Smile and kill".
Catchy, isn't it? "Do something slightly different and give it a totally new
name if you want to catch the attention of community", is what is happening in
biology these days (if one among them is to be believed.).
Would love to read something about how text (if not information) is retrieved
from the large corpus these days. Would anyone from that course you mentioned be
interested to write down an introduction?
--
Dilawar
NCBS Bangalore
EE, IIT Bombay
>> hands here<https://docs.google.com/forms/d/1JabDR06INIdG4ZFnJNaAYDXIPs8CWK3quJ5waK4M5ic/viewform>
>> . *
algorithm", though technically one can call a collection of algorithms "an
algorithm".
What happens during search these days? I am guessing; do a search the way you
have been doing so far, then find a pattern in corpus using, let say simulating
annealing; check if the result make sense, if not do a machine learning and
build a database; etc. etc..
And when product is ready, give it a nice name. This trick is quite popular
these days. Intel announced "3D transistors" as a new keyword. Now which
transistor is not a 3D transistor.
A few days ago, there was a talk here on a "new" approach to remove extra
terrestrial pathogens from human body; the approach was titled "Smile and kill".
Catchy, isn't it? "Do something slightly different and give it a totally new
name if you want to catch the attention of community", is what is happening in
biology these days (if one among them is to be believed.).
Would love to read something about how text (if not information) is retrieved
from the large corpus these days. Would anyone from that course you mentioned be
interested to write down an introduction?
--
Dilawar
NCBS Bangalore
EE, IIT Bombay
On Sun, Oct 06, 2013 at 01:19:42AM +0530, Nivvedan (நிவேதன்) wrote:
>+1
>
>This email was sent from a device/client/interface that unfortunately
>cannot sign emails.
>
>Trust the authenticity of this message at your own peril. If
>doubtful/paranoid, request me for a signed email and I shall get back to
>you as soon as I can.
>
>
>2013/10/5 Guna Prasaad <guna...@gmail.com>
>
>> Last week Google announced a brand new algorithm for its search engine,
>> called *Humming Bird*. Although Google has not given away any details, it
>+1
>
>This email was sent from a device/client/interface that unfortunately
>cannot sign emails.
>
>Trust the authenticity of this message at your own peril. If
>doubtful/paranoid, request me for a signed email and I shall get back to
>you as soon as I can.
>
>
>2013/10/5 Guna Prasaad <guna...@gmail.com>
>
>> Last week Google announced a brand new algorithm for its search engine,
>> has said that the new algorithm focuses on ranking information based on a
>> better understanding of search queries.
>>
>> The search results we currently receive reflect the matching combination
>> of key words that a search phrase contains, rather than the true meaning of
>> the sentence itself. Search results produced by Hummingbird will reflect
>> the full semantic meaning of longer search phrases, and should in theory
>> produce more accurate results. For example Hummingbird will more greatly
>> consider question words like “how” “why”, “where” and “when” in search
>> phrases, in addition to content keywords. Hence Hummingbird moves the
>> emphasis of search from “results” to “answers”. The semantic search
>> capabilities of Hummingbird aim to address this need.
>>
>> This is something that I could get by random web scraping from blogs on
>> humming bird algorithm. I think this is an interesting topic and many would
>> like to know more about this, especially people who are enrolled in CS635 (Information
>> Retrieval & Mining for Hypertext & the Web). *Lets have an informal
>> better understanding of search queries.
>>
>> The search results we currently receive reflect the matching combination
>> of key words that a search phrase contains, rather than the true meaning of
>> the sentence itself. Search results produced by Hummingbird will reflect
>> the full semantic meaning of longer search phrases, and should in theory
>> produce more accurate results. For example Hummingbird will more greatly
>> consider question words like “how” “why”, “where” and “when” in search
>> phrases, in addition to content keywords. Hence Hummingbird moves the
>> emphasis of search from “results” to “answers”. The semantic search
>> capabilities of Hummingbird aim to address this need.
>>
>> This is something that I could get by random web scraping from blogs on
>> humming bird algorithm. I think this is an interesting topic and many would
>> like to know more about this, especially people who are enrolled in CS635 (Information
>> meetup on this topic, where we can have a discussion on this, based on
>> whatever information each one could gather. **J**ust a small show of
>> hands here<https://docs.google.com/forms/d/1JabDR06INIdG4ZFnJNaAYDXIPs8CWK3quJ5waK4M5ic/viewform>
>> . *
Mayank Singhal
Oct 6, 2013, 5:05:51 AM10/6/13
to wncc...@googlegroups.com
> Intel announced "3D transistors" as a new keyword
Aren't 3D transistors actually more three dimensional than their predecessors in the sense that the silicon part is more than just a 2D layer? If I remember correctly, they created a new fabrication process that allowed the silicon part (I don't even remember what these things are called) to have teeth increase contact area or something.>> email to wncc_iitb+unsubscribe@googlegroups.com.
>> For more options, visit https://groups.google.com/groups/opt_out.
>>
>
>--
>--
>The website for the club is http://stab-iitb.org/wncc
>To post to this group, send email to wncc...@googlegroups.com
>
>---
>You received this message because you are subscribed to the Google Groups "Web and Coding Club IIT Bombay" group.
>To unsubscribe from this group and stop receiving emails from it, send an email to wncc_iitb+unsubscribe@googlegroups.com.
>For more options, visit https://groups.google.com/groups/opt_out.
--
--
The website for the club is http://stab-iitb.org/wncc
To post to this group, send email to wncc...@googlegroups.com
---
You received this message because you are subscribed to the Google Groups "Web and Coding Club IIT Bombay" group.
To unsubscribe from this group and stop receiving emails from it, send an email to wncc_iitb+unsubscribe@googlegroups.com.
Manish Goregaokar
Oct 6, 2013, 6:08:30 AM10/6/13
to wncc...@googlegroups.com
This is probably another augmentation of their algorithm from their large databases, statistical machine learning style. I don't think that it's really a new algorithm, just that their gradual layering of the older algorithms has reached a point where they give it a new name. I'd have to read more about this to be sure though.
-Manish Goregaokar
To unsubscribe from this group and stop receiving emails from it, send an email to wncc_iitb+...@googlegroups.com.
Dilawar Singh
Oct 6, 2013, 6:13:30 AM10/6/13
to wncc...@googlegroups.com
Well, as far as degree of 3d'ness is concerned, yes and no. There is very little
addition to 3-d'ness as far as chip is concerned unless you ignore the substrate
in which it is embedded. Even if you ignore the substrate, following is the
argument.
They have modified the structure of the gate. In standard device, there is one
gate which controls the flow of electron (or holes) by enhancing or depleting
the channel. To draw an analogy, gate is just like the knob in the tap
controlling the flow of water.
Gate can also be seen as a capacitor. And if switching has to happen fast, you
need better control over this channel (how fast you can charge it) and also how
strong this capacitor is to induce an effect in the channel. Process is slightly
new, because intel replace SiO2 with Hafnium Oxide as the insulator between gate
and channel and that was something very new. After this, it is more like, Ok now
what else we can do.
It turns out, instead of one gate (one large capacitor), you can get better
switching by three gates (three smaller capacitors in parallel) over the same
length of gate. A competitive technology ( perhaps a generalisation of it) is
called fin-FET but fabricating millions of transistor over the chip using this
technology has a variation problem.
Largest contributors to 3-dness of a chip are metal layers which are usually 12
or more and via are used to connect layers to layer (just imagine a 12 layer
PCB). Now what intel has done, as far as degree of 3D-ness is concerned, is add
3 more rooms in a floor of 12 story building. Its not like the full circuit is
anyway more 3D but this was not your argument anyway.
I dont have model to measure a 3-D'ness of a single transistor right away so
lets just check the height of overall device. Ignore the substrate. The source
and drain junctions are 10-15 um deep (And thats a big number, this is the best
one can do for 22nm tech). Over a channel, there is oxide which is 5-6 molecular
layer (7-9 A). On top of it has a gate whose height is (varies a lot because you
have to make a metal connection to it), approx 30-100 um. Intel may be on the
lower side. This structure is over the substrate, like a bench on the ground.
Now the point is, if a bench of width d and length l and height h, is less 3-D
than three benches of width d/3, length l, and height h, kept in parallel
separated by, assume d/3? But that also increase the channel length. So yes, if
your model of 3D-ness takes this into account. Mine does not, one bench is as 3D
as 3 of them.
The illusion of planarity is due to the fact the source and drain junctions are
buried inside the substrate. But gate is anyway fabricated on the top: whether
one or 3.
--
Dilawar
NCBS Bangalore
EE, IIT Bombay
>> >> hands here<https://docs.google.com/**forms/d/**
>> 1JabDR06INIdG4ZFnJNaAYDXIPs8CW**K3quJ5waK4M5ic/viewform<https://docs.google.com/forms/d/1JabDR06INIdG4ZFnJNaAYDXIPs8CWK3quJ5waK4M5ic/viewform>
>> .
>> >> For more options, visit https://groups.google.com/**groups/opt_out<https://groups.google.com/groups/opt_out>
>> .
>> .
>> >For more options, visit https://groups.google.com/**groups/opt_out<https://groups.google.com/groups/opt_out>
>> .
>> .
>> For more options, visit https://groups.google.com/**groups/opt_out<https://groups.google.com/groups/opt_out>
>> .
addition to 3-d'ness as far as chip is concerned unless you ignore the substrate
in which it is embedded. Even if you ignore the substrate, following is the
argument.
They have modified the structure of the gate. In standard device, there is one
gate which controls the flow of electron (or holes) by enhancing or depleting
the channel. To draw an analogy, gate is just like the knob in the tap
controlling the flow of water.
Gate can also be seen as a capacitor. And if switching has to happen fast, you
need better control over this channel (how fast you can charge it) and also how
strong this capacitor is to induce an effect in the channel. Process is slightly
new, because intel replace SiO2 with Hafnium Oxide as the insulator between gate
and channel and that was something very new. After this, it is more like, Ok now
what else we can do.
It turns out, instead of one gate (one large capacitor), you can get better
switching by three gates (three smaller capacitors in parallel) over the same
length of gate. A competitive technology ( perhaps a generalisation of it) is
called fin-FET but fabricating millions of transistor over the chip using this
technology has a variation problem.
Largest contributors to 3-dness of a chip are metal layers which are usually 12
or more and via are used to connect layers to layer (just imagine a 12 layer
PCB). Now what intel has done, as far as degree of 3D-ness is concerned, is add
3 more rooms in a floor of 12 story building. Its not like the full circuit is
anyway more 3D but this was not your argument anyway.
I dont have model to measure a 3-D'ness of a single transistor right away so
lets just check the height of overall device. Ignore the substrate. The source
and drain junctions are 10-15 um deep (And thats a big number, this is the best
one can do for 22nm tech). Over a channel, there is oxide which is 5-6 molecular
layer (7-9 A). On top of it has a gate whose height is (varies a lot because you
have to make a metal connection to it), approx 30-100 um. Intel may be on the
lower side. This structure is over the substrate, like a bench on the ground.
Now the point is, if a bench of width d and length l and height h, is less 3-D
than three benches of width d/3, length l, and height h, kept in parallel
separated by, assume d/3? But that also increase the channel length. So yes, if
your model of 3D-ness takes this into account. Mine does not, one bench is as 3D
as 3 of them.
The illusion of planarity is due to the fact the source and drain junctions are
buried inside the substrate. But gate is anyway fabricated on the top: whether
one or 3.
--
Dilawar
NCBS Bangalore
EE, IIT Bombay
On Sun, Oct 06, 2013 at 09:05:51AM +0000, Mayank Singhal wrote:
>> Intel announced "3D transistors" as a new keyword
>
>Aren't 3D transistors actually more *three dimensional *than their
>> Intel announced "3D transistors" as a new keyword
>
>> 1JabDR06INIdG4ZFnJNaAYDXIPs8CW**K3quJ5waK4M5ic/viewform<https://docs.google.com/forms/d/1JabDR06INIdG4ZFnJNaAYDXIPs8CWK3quJ5waK4M5ic/viewform>
>> >
>> >> . *
>> >>
>> >>
>> >> --
>> >> --
>> >> The website for the club is http://stab-iitb.org/wncc
>> >> To post to this group, send email to wncc...@googlegroups.com
>> >>
>> >> ---
>> >> You received this message because you are subscribed to the Google
>> Groups
>> >> "Web and Coding Club IIT Bombay" group.
>> >> To unsubscribe from this group and stop receiving emails from it, send
>> an
>> >> email to wncc_iitb+unsubscribe@**googlegroups.com<wncc_iitb%2Bunsu...@googlegroups.com>
>> >> . *
>> >>
>> >>
>> >> --
>> >> --
>> >> The website for the club is http://stab-iitb.org/wncc
>> >> To post to this group, send email to wncc...@googlegroups.com
>> >>
>> >> ---
>> >> You received this message because you are subscribed to the Google
>> Groups
>> >> "Web and Coding Club IIT Bombay" group.
>> >> To unsubscribe from this group and stop receiving emails from it, send
>> an
>> .
>> >> For more options, visit https://groups.google.com/**groups/opt_out<https://groups.google.com/groups/opt_out>
>> .
>> >>
>> >
>> >--
>> >--
>> >The website for the club is http://stab-iitb.org/wncc
>> >To post to this group, send email to wncc...@googlegroups.com
>> >
>> >---
>> >You received this message because you are subscribed to the Google Groups
>> "Web and Coding Club IIT Bombay" group.
>> >To unsubscribe from this group and stop receiving emails from it, send an
>> email to wncc_iitb+unsubscribe@**googlegroups.com<wncc_iitb%2Bunsu...@googlegroups.com>
>> >
>> >--
>> >--
>> >The website for the club is http://stab-iitb.org/wncc
>> >To post to this group, send email to wncc...@googlegroups.com
>> >
>> >---
>> >You received this message because you are subscribed to the Google Groups
>> "Web and Coding Club IIT Bombay" group.
>> >To unsubscribe from this group and stop receiving emails from it, send an
>> .
>> >For more options, visit https://groups.google.com/**groups/opt_out<https://groups.google.com/groups/opt_out>
>> .
>>
>> --
>> --
>> The website for the club is http://stab-iitb.org/wncc
>> To post to this group, send email to wncc...@googlegroups.com
>>
>> ---
>> You received this message because you are subscribed to the Google Groups
>> "Web and Coding Club IIT Bombay" group.
>> To unsubscribe from this group and stop receiving emails from it, send an
>> email to wncc_iitb+unsubscribe@**googlegroups.com<wncc_iitb%2Bunsu...@googlegroups.com>
>> --
>> --
>> The website for the club is http://stab-iitb.org/wncc
>> To post to this group, send email to wncc...@googlegroups.com
>>
>> ---
>> You received this message because you are subscribed to the Google Groups
>> "Web and Coding Club IIT Bombay" group.
>> To unsubscribe from this group and stop receiving emails from it, send an
>> .
>> For more options, visit https://groups.google.com/**groups/opt_out<https://groups.google.com/groups/opt_out>
>> .
>>
>
>--
>--
>The website for the club is http://stab-iitb.org/wncc
>To post to this group, send email to wncc...@googlegroups.com
>
>---
>You received this message because you are subscribed to the Google Groups "Web and Coding Club IIT Bombay" group.
>To unsubscribe from this group and stop receiving emails from it, send an email to wncc_iitb+...@googlegroups.com.
>
>--
>--
>The website for the club is http://stab-iitb.org/wncc
>To post to this group, send email to wncc...@googlegroups.com
>
>---
>You received this message because you are subscribed to the Google Groups "Web and Coding Club IIT Bombay" group.
Dilawar Singh
Oct 6, 2013, 6:50:32 AM10/6/13
to wncc...@googlegroups.com
Sorry I was wrong about a key change. I got confused with tri-Gate and 3-D
transistors used for same device by intel. The process document shows that
channel is not buried inside the substrate as is the case with usual device. So
yes, there is an addition of three bench like structures but geometrically, the
argument about 3-D'ness is still valid.
The process used for raising the channel outside the substrate is new. The
distinction they are making: not layer by layer. Thats novel but, more 3D?
--
Dilawar
NCBS Bangalore
EE, IIT Bombay
>> >> hands here<https://docs.google.com/**forms/d/**
>> 1JabDR06INIdG4ZFnJNaAYDXIPs8CW**K3quJ5waK4M5ic/viewform<https://docs.google.com/forms/d/1JabDR06INIdG4ZFnJNaAYDXIPs8CWK3quJ5waK4M5ic/viewform>
>> .
>> >> For more options, visit https://groups.google.com/**groups/opt_out<https://groups.google.com/groups/opt_out>
>> .
>> .
>> >For more options, visit https://groups.google.com/**groups/opt_out<https://groups.google.com/groups/opt_out>
>> .
>> .
>> For more options, visit https://groups.google.com/**groups/opt_out<https://groups.google.com/groups/opt_out>
>> .
transistors used for same device by intel. The process document shows that
channel is not buried inside the substrate as is the case with usual device. So
yes, there is an addition of three bench like structures but geometrically, the
argument about 3-D'ness is still valid.
The process used for raising the channel outside the substrate is new. The
distinction they are making: not layer by layer. Thats novel but, more 3D?
--
Dilawar
NCBS Bangalore
EE, IIT Bombay
On Sun, Oct 06, 2013 at 09:05:51AM +0000, Mayank Singhal wrote:
>> Intel announced "3D transistors" as a new keyword
>
>Aren't 3D transistors actually more *three dimensional *than their
>
>> 1JabDR06INIdG4ZFnJNaAYDXIPs8CW**K3quJ5waK4M5ic/viewform<https://docs.google.com/forms/d/1JabDR06INIdG4ZFnJNaAYDXIPs8CWK3quJ5waK4M5ic/viewform>
>> >
>> >> . *
>> >>
>> >>
>> >> --
>> >> --
>> >> The website for the club is http://stab-iitb.org/wncc
>> >> To post to this group, send email to wncc...@googlegroups.com
>> >>
>> >> ---
>> >> You received this message because you are subscribed to the Google
>> Groups
>> >> "Web and Coding Club IIT Bombay" group.
>> >> To unsubscribe from this group and stop receiving emails from it, send
>> an
>> >> email to wncc_iitb+unsubscribe@**googlegroups.com<wncc_iitb%2Bunsu...@googlegroups.com>
>> >> . *
>> >>
>> >>
>> >> --
>> >> --
>> >> The website for the club is http://stab-iitb.org/wncc
>> >> To post to this group, send email to wncc...@googlegroups.com
>> >>
>> >> ---
>> >> You received this message because you are subscribed to the Google
>> Groups
>> >> "Web and Coding Club IIT Bombay" group.
>> >> To unsubscribe from this group and stop receiving emails from it, send
>> an
>> .
>> >> For more options, visit https://groups.google.com/**groups/opt_out<https://groups.google.com/groups/opt_out>
>> .
>> >>
>> >
>> >--
>> >--
>> >The website for the club is http://stab-iitb.org/wncc
>> >To post to this group, send email to wncc...@googlegroups.com
>> >
>> >---
>> >You received this message because you are subscribed to the Google Groups
>> "Web and Coding Club IIT Bombay" group.
>> >To unsubscribe from this group and stop receiving emails from it, send an
>> email to wncc_iitb+unsubscribe@**googlegroups.com<wncc_iitb%2Bunsu...@googlegroups.com>
>> >
>> >--
>> >--
>> >The website for the club is http://stab-iitb.org/wncc
>> >To post to this group, send email to wncc...@googlegroups.com
>> >
>> >---
>> >You received this message because you are subscribed to the Google Groups
>> "Web and Coding Club IIT Bombay" group.
>> >To unsubscribe from this group and stop receiving emails from it, send an
>> .
>> >For more options, visit https://groups.google.com/**groups/opt_out<https://groups.google.com/groups/opt_out>
>> .
>>
>> --
>> --
>> The website for the club is http://stab-iitb.org/wncc
>> To post to this group, send email to wncc...@googlegroups.com
>>
>> ---
>> You received this message because you are subscribed to the Google Groups
>> "Web and Coding Club IIT Bombay" group.
>> To unsubscribe from this group and stop receiving emails from it, send an
>> email to wncc_iitb+unsubscribe@**googlegroups.com<wncc_iitb%2Bunsu...@googlegroups.com>
>> --
>> --
>> The website for the club is http://stab-iitb.org/wncc
>> To post to this group, send email to wncc...@googlegroups.com
>>
>> ---
>> You received this message because you are subscribed to the Google Groups
>> "Web and Coding Club IIT Bombay" group.
>> To unsubscribe from this group and stop receiving emails from it, send an
>> .
>> For more options, visit https://groups.google.com/**groups/opt_out<https://groups.google.com/groups/opt_out>
>> .
>>
>
>--
>--
>The website for the club is http://stab-iitb.org/wncc
>To post to this group, send email to wncc...@googlegroups.com
>
>---
>You received this message because you are subscribed to the Google Groups "Web and Coding Club IIT Bombay" group.
>To unsubscribe from this group and stop receiving emails from it, send an email to wncc_iitb+...@googlegroups.com.
>
>--
>--
>The website for the club is http://stab-iitb.org/wncc
>To post to this group, send email to wncc...@googlegroups.com
>
>---
>You received this message because you are subscribed to the Google Groups "Web and Coding Club IIT Bombay" group.
Reply all
Reply to author
Forward
0 new messages