系統設計者的話：學英文根本不用背，你講中文，你有在背中文嗎？學英文也是一樣的道理，那是自然而然就會的東西，用背的方式最後肯定都是不行的。 更多...

最好的學習就是有英文的父母，不斷的跟你說話、耐心的教導你，時間長了，你也就慢慢就融入其中了。

但是我們沒有英文父母怎麼辦？看影片學英文就學的起來嗎？學不起來。

我們從小到大，也看過了不少英文電影，至少也看過1、200部也跑不掉，但是為什麼看了這麼多的英文電影，英文還是很爛呢？

主要的原因是我們沒有從基礎性的強化學習。舉個例子，西方的幼兒出生時，他至少聽了各種大人講英文至少聽了2年，才慢慢開始說英文、與大人互動。

而且都是一字一句慢慢的說、慢慢的教、漸進式的、慢慢的、越來越習慣、越來越聽的懂、越來越會說的過程。

那麼我們看的影片、電影過程中，都是大人成熟型的講話速度、方式、語調，我們怎麼可能跟的上？

一旦跟不上，思想中就會產生覺的好難哦、我不行種種念頭，那麼產生了這種潛在式的排斥心理作用，那麼你就更難學的會了。


所以我們必須打好基礎、深厚的基礎，萬丈高樓平地起，唯有打下深厚的基礎，後面你走的路才會越來越順。

程式界裏有句話，程式債欠的越多，最後就只能放棄。學英文也是一樣，英文單字債欠的越多，你就直接上來文法、句型、語調，最後也只能放棄。

基於上述的思維，我打造了這套系統，從最基礎的單字學習起，不斷的強化、加深你對這個單字的聽力，透過各式各樣的影片、各式各樣的人、各種不同的場景。

就只是為了訓練一個單字的聽力，無數的人跟你講同一個單字，講個20遍、50篇、100遍(且不重覆)，我就不信學不起來。


這套系統做出來後，我自己在使用的過程中，確實真的能聽出來他們講的這個單字，而且他們講的速度再快也能聽的出來。

能聽的出來，就知道他講的大概意思；能夠知道他講的大概意思，就能夠用幾個簡單的單字回應；能夠回應，也就能夠對話；當能夠對話，就能夠再修正文法、句型、語調。


我的目地就是希望自己至少能聽的懂人家在說什麼，用英文跟人家對話，大家共同努力、學習。若有其它想法或對這套系統設計的建議，也可以與我聯絡：ichich2013@gmail.com

Distributed TensorFlow (TensorFlow @ O’Reilly AI Conference, San Francisco '18)
觀看次數：373
devices
such
as
cpus
gpus
or
gpus
to
losses
and
gradients
get
aggregated
box
ideally
the
user
shouldn't
have
to
architectures
like
synchronous
or
core
cpu
on
your
desktop
tensorflow
cpu
automatically
next
you
may
add
a
gpu
tensorflow
will
automatically
run
your
recommended
tensorflow
high-level
api
we
reuse
the
resinate
50
that's
prepackaged
an
optimizer
that
we'll
be
using
in
our
components
of
tensor
code
to
be
optimizer
batch
norm
layers
metrics
and
parallelism
let's
say
you
have
n
workers
or
n
devices
in
data
parallelism
each
computation
but
with
a
different
subset
replicated
as
a
0
a
1
a
2
and
a
3
across
these
four
variables
conceptually
form
mirrored
variable
these
variables
are
on
each
device
it's
a
fuse
algorithm
different
devices
one
common
algorithm
network
bandwidth
optimally
because
it
implementations
of
all
reduce
for
vendors
such
as
the
nv
nvidia
provide
synchronous
training
so
let's
see
how
variables
are
replicated
across
the
two
variables
it
then
computes
a
backward
pass
and
computes
the
gradients
once
the
gradients
are
computed
on
each
gradients
and
once
the
grains
are
the
combined
gradients
and
updated
its
training
time
by
overlapping
computation
of
gradients
at
lower
layers
in
the
layer
a
while
you're
transmitting
the
gpus
we
use
a
resonate
50
model
with
it's
a
very
popular
benchmark
for
nvidia
tesla
B
100
gpus
on
Google
cloud
garden
repo
and
currently
it
uses
the
estimator
API
we're
working
on
cara's
cloudy
pews
that
are
interconnected
to
estimator
API
using
tpu
strategy
so
via
fast
interconnects
and
this
means
malta
is
these
overheads
so
setting
this
your
pods
consists
of
many
cloud
cpus
interconnected
via
fast
Network
and
this
on
to
multi
node
with
gpus
thank
you
all
multi-gpu
multi
node
support
we
have
currently
support
for
pre-made
estimators
in
tensorflow
111
which
is
estimator
to
model
estimator
to
model
s
model
to
estimator
tvrs
estimator
model
to
estimator
that
takes
a
karass
model
technology
that's
open-source
now
called
kubernetes
as
the
orchestration
platform
to
fire
up
multiple
nodes
and
kubernetes
called
tf
underscore
config
they've
the
hosts
that
participates
in
the
so
that's
how
you
need
to
configure
your
configure
this
kubernetes
provides
an
tensorflow
well
it's
part
of
the
initial
jinja
tensorflow
tensorflow
repository
this
is
generate
automatically
generate
the
set
the
worker
replicas
to
three
nodes
macro
that
populates
tf
config
based
on
we've
now
configured
the
kubernetes
is
the
estimator
lingo
so
I
provide
a
config
here
you
see
the
wrong
config
reduced
strategy
so
not
mirrored
have
multi-gpu
multi
node
multi-gpu
in
each
of
the
nodes
configured
for
tensorflow
ok
and
so
let's
look
at
this
on
mirrored
strategy
but
it
adds
functionality
in
order
to
deal
with
that
it
automatically
selects
the
best
they
are
traversing
hosts
with
all
of
doing
that
depending
on
the
interconnect
between
these
hosts
another
algorithm
is
characteristics
of
your
setup
and
automatically
detect
the
best
algorithm
so
that
was
multi
node
multi
accelerator
so
what
i'm
describing
here
is
worker
nodes
and
they're
working
on
batches
of
this
shared
storage
they're
are
working
independently
calculating
variables
down
to
the
workers
so
it's
just
specifying
the
worker
replicas
in
underscore
replicas
so
that's
the
number
kubernetes
setup
now
what
about
the
code
you
saw
the
run
config
the
confi
worker
you
specify
the
config
object--
it
will
automatically
distribute
the
configurations
tensorflow
multi
node
training
so
let's
summarize
kubernetes
to
distribute
tensor
flow
compiles
or
translates
over
to
the
jama
find
tensorflow
that
contrib
dot
with
single
node
multi-gpu
we've
continuously
working
on
trying
to
configurations
we've
talked
about
tpu
in
kara's
native
kara's
code
we
can
important
a
feature
of
tensorflow
so
if
synchronous
synchronous
gradient
updates
between
1
and
8
gpus
on
a
kubernetes
scale
across
gpus
different
accelerators
well
as
multi
node
and
tpu
pod
right
and

相關影片

加Line好友 每日8AM 推播1單字

加入好友

其他人正在學習的單字

1. all
2024-05-02 20:24
2. website
2024-05-02 20:24
3. join
2024-05-02 20:24
4. kick
2024-05-02 20:24
5. bring
2024-05-02 20:24
6. kit
2024-05-02 20:24
7. they
2024-05-02 20:24
8. invaded
2024-05-02 20:24
9. juice
2024-05-02 20:24
10. tracks
2024-05-02 20:24
11. hit
2024-05-02 20:24
12. momentum
2024-05-02 20:24
13. scale
2024-05-02 20:24
14. it
2024-05-02 20:24
15. full
2024-05-02 20:23

本月觀看影片單字排行榜

本月查閱單字排行榜

本月聆聽單字排行榜