Thứ Sáu, 3 tháng 8, 2018

Waching daily Aug 3 2018

Lingke Family Summer camps oversized swim ring and Rowing competition challenge! | Xiaoling toys

For more infomation >> Lingke Family Summer camps oversized swim ring and Rowing competition challenge! | Xiaoling toys - Duration: 8:51.

-------------------------------------------

বিদআতি চেনার সহজ উপায় || Bidati Chinar sohoj upai || Motiur Rahman Madani || Bangla Waz short Video - Duration: 5:09.

For more infomation >> বিদআতি চেনার সহজ উপায় || Bidati Chinar sohoj upai || Motiur Rahman Madani || Bangla Waz short Video - Duration: 5:09.

-------------------------------------------

اختراق جهاز لمجندة صهيونية ..صقر الجزائر - Duration: 6:40.

For more infomation >> اختراق جهاز لمجندة صهيونية ..صقر الجزائر - Duration: 6:40.

-------------------------------------------

Failing at Machine Learning (Blocky part 2) - Pwn Adventure 3 - Duration: 14:34.

Let's continue with the Blocky puzzle challenge.

We identified a packet sent from the server to the client that contains the state of the

input buttons, 32bits and the state of all the output lines.

We then used the proxy to inject random input states and collect a lot of data.

And so I have been running this brute-forcing over night.

wc -l counts the lines of a file and with watch, which is just executing wc over and

over, we can easily observe how the test data grows.

So shortly after I started around midnight I had already 4000 samples, and then I went

to bed.

In the next morning after 9:00 am we had 100.000 samples.

But I let it run even longer and in the end I had about 340.000 samples.

So… how can we do machine learning with that?

First of all, before we even do any real analysis of this data, we have to figure out which

output bits are actually important for us.

Not all of the over 170 bits are useful.

Some are just the lines directly connected to the button, or other lines in the middle.

So we have to figure out which are the important output bits.

If you look at the data you can immediately see that every output starts with 6 1s, and

then two 0s.

And that makes sense if you look at the end of the puzzle.

The final stage of the puzzle just combines a lot of output lines with OR gates.

So if only one of these lines is HIGH, a 1, then the OR result will be HIGH, and all these

lines will be high.

You can see for example here the very first OR gate in this chain, with 3 inputs.

And with the random button inputs it's very rare that all 3 of them are 0.

Thus the output is almost always high and thus the whole chain stays high.

And at the end of the chain the signal is inverted, and so we then have two low.

Because we probably never guessed the correct input and one of these lines will always be

high, thus we see this pattern here.

This means that the important output lines for us are the ones feeding into the final

OR gates.

Those must be 0, and then the whole line becomes 0 and thus the final door opens.

Now we need to find which bits in our output correspond to these OR inputs.

To do that I simply position myself over the puzzle and then use the proxy to inject a

random input.

Then I write down the state of one OR input.

Get another random input, and write down the states.

And I'm continuing this for a bit and we get this ON/OFF pattern for these two lines

to the OR gate.

Now we can take our over 170 bits output and try to find this pattern in there.

To help with that I wrote a few simple loops in python.

Btw I'm using jupyter notebook here, easy to run with Anaconda, because this kind of

exploratory data analysis is really nice with it.

The outer loop goes over each bit.

So first we look at the first bit of our output.

Then we go over each test case.

So we basically compare the first bit of the output with the pattern we have written down.

If they match, cool.

But if they don't match we break out of the loop.

And then we check the next bit.

We go over each second bit in the output, compare it to our pattern and if it finds

a bit that doesn't match, we break out.

But if all bits would match, we would never break out of the for loop and then this else

case of the for loop is executed.

YEP, for loops can have an else cases, which is executed when we didn't break.

This means we found a output bit that matches the pattern.

But when we did that we find two matching bits.

172 and 173.

Let's see if they indeed are always the same, or if we just need more data.

So I collect a few more samples so that if indeed it was just coincidence, then the probability

that they would still match goes very low.

And when we run this now, indeed, we see now that 173 is the magic bit.

Now we just have to keep doing that for all OR gate inputs.

Write it down and then we get all the important output bits.

Infact there are 15.

So we can ignore the rest.

So our goal is it to get all of these bits to 0.

Because then the whole chain stays off, the inverter makes it high and the door will open.

Before I mentioned anything about machine learning, I wanted to say that what I have

is "gefährliches halbwissen".

That is a cool german expression and it translates to "dangerous superficial knowledge",

you use this to describe somebody who knows a little bit and they think they know enough

to voice an opinion, but their views or opinions are dangerous because they only have superficial

knowledge and don't really get it.

And that very much applies to me.

So I will not prefix everything I say with a "sorry, maybe, I'm not sure", and

will just tell you what is currently going on in my head.

I'm just sharing with you the process of me learning.

And that also means to do dumb stuff.

So I try to use the little bit of knowledge I have and see how far I can get.

I know very well that a professional would do it completely differently.

And in some way I'm really scared to share my thoughts, that's also the reason for

this disclaimer, because I'm well aware that they are probably wrong.

So if you are a professional in this field, I'd love to know what I got right and what

I got wrong.

And if somebody of you can make a proper model, I would LOVE to see it.

So first of all, what kind of algorithm or method of machine learning or AI should we

used.

Most of you probably know about neural networks, deep learning and stuff.

And we have 32 input bits and 15 important output bits.

That could be something that could be modeled with a neural network, right?

Each input bit connects to simple binary functions, ANDs and ORs and those outputs connect again

with other simple functions and eventually result in the 15 output bits.

And I think a well modelled neural network could totally solve this.

Basically each neuron, or a combination of neurons could implement such a simple logical

function, and thus if we have multiple layers it should theoretically be perfectly able

to learn that.

But if you model it like that, then you are just predicting the output for a given input.

And that doesn't really help us because we don't want to know what a given input

gives us, we can just test that already ingame, right.

What we want is, we know the output we want.

We want the output to be all 0.

So actually what we want to know is, what inputs lead to the outputs being all 0.

Which means we probably want to reverse the model.

The output bits become the input of the network and the input bits become the output of the

neural network.

So the neural network would train the theoretical inverse of the binary function.

Given these output bits, what was the input for it.

If that works, we could maybe ask the network to tell us, here we have these 0 output bits,

what would the input be.

But the inverse of an OR and AND function is not that simple, right?

For example for an OR gate you know that 0 means the inputs were 0, but if it was 1,

then one or both inputs could have been 1.

So that introduces an uncertainty and I would hope the neural network still could work with

that, but I don't know.

The whole magic or difficulty for neural networks, as far as I understand, is mostly deciding

how many layers the network should have, what kind of activation function to use and a lot

of other stuff.

And that's exactly why I have no clue what I am doing.

I think I could look at the activation functions and know that for our case we probably don't

want something like sigmoid because that models more like analog data, and we want something

more digital, ON/OFF, so maybe a Rectified linear unit or even binary step function,

because here certain connections can be completely turned off - and that's what we want because

not every input bit is connected to every function in the real network.

I even went so far to checkout tensorflow, I installed it and I googled for "neural

networks boolean tensorflow", because essentially we have boolean data.

And there was one blog post in particular, "Solving XOR with a Neural Network in Tensorflow",

which caught my attention, because we also want to learn boolean functions, just something

more complex..

I and I started to adapt that code to my problem, but then I also kinda gave up…

I think I need to do a bit more learning of basics and playing with known simpler examples

before I try to apply this to something so weird.

But there was another thing I wanted to try out.

You see neural networks are just one topic, there are a lot of other machine learning

algorithms out there.

And there is a whole subcategory about learning rules and decision trees.

The goal [of a decision tree] is to create a model that predicts the value of a target

variable based on several input variables.

That's exactly what we want.

I thought maybe we can just use such an algorithm to find exactly the binary AND and OR combinations

of input bits that lead to this particular output.

And then we can solve that with z3 And I remembered a tool that I have seen once

called weka.

Weka is a collection of machine learning algorithms for data mining tasks.

It's a tool from the Machine Learning Group at the University of Waikato in New Zealand.

To work with this tool I quickly transform the data into an arff file format, which basically

just describes the attributes we have in our file and splits it all to comma separated

values.

So here we can open the file and load all attributes.

Let's only look at one particular output bit, so I remove the others from the dataset.

Then in the classify tab we can select an algorithm from a list of many many different

ones.

Again, I have no clue what I am doing so I just select one, let it run and see what happens.

As a test the output bit I selected is not one of our final ones, it's a simpler one,

an output state much earlier in the whole puzzle.

There are also a lot of variables you can control, but I have no clue so I just click

Start.

And very quickly it extracted some rules, a decision list.

So this particular output bit is influenced by input bit_31 and bit_12.

Let's write this down in a table.

It basically says, if input bit_31 was 0, then the output bit_17 is 0 as well.

Which means it doesn't matter if bit_12 was 0 or 1 But if the input bit_31 was not

0, so it was 1, then bit_12 is important.

If bit_12 is 1, then our output is also 0, but if that was not the case, if bit_12 was

0, then the output bit is 1.

If we draw this as gates we have an AND gate that takes in the bit_31, but only triggers

when bit_12 is 0, so bit_12 goes through an inverter first.

So now only if bit_31 is high and bit_12 is low, the AND gate will put out a 1.

Cool!

So we used a machine learning algorithm called PART to learn a simple gate.

Weka will also run this model on other data in the dataset and you see in all testcases

this ruleset is perfect.

100% accuracy.

So that is really cool and I tried it with one of our final output bits, which have much

more complex connections and gates going on.

That runs a bit longer and when finally a model was found, then it's just insane long.

It's huuuuge.

The number in parentheses shows how many of testcases are covered up to this particular

rule.

For the simple gate before half of all cases were already covered by the first rule, and

the other half was handled by these two rules.

But here we have rules that sometimes have only like 6 or 3 cases in our dataset.

So that doesn't look good.

That looks like overfitting.

And also the test on the whole dataset reveals, that it's only 99.8% correct.

Which is not bad, for a lot of real life applications it's probably really good, but I was hoping

such an algorithm could really uncover the logic connections perfectly.

I tried a couple of different algorithms and weka really makes this explorative process

fun.

But again my lack of experience probably fails me here.

Maybe one of these algorithms if the variables are tweaked correctly could be perfect.

But I don't know.

I gave up on this too.

Now you see I did all this over the course of many many hours and so I was handling this

training data a lot.

And one time I thought I could checkout how many inputs actually I resulted in a desired

0 in one of these output bits and I wrote this simple loop.

Just print the input if the particular output bit was 0.

I ran this and I couldn't believe my eyes.

While the 0s and 1s were rushing by, I noticed one column to just stayed constant.

Look at this!

And that turned out to be true for every output bit.

Each output bit had at least one bit constant in the input.

Now the inverse of this was not true, not every input where this was a 1 resulted in

this output to be 0, but in every case where it was 0, this input bit was 1.

This means, our final solution MUST have these bits set to this constant value.

So I wrote a quick script to extract all constant bits and this is what I got.

This really surprised me.

Every second bit was constant!

Holy shit!

This is awesome.

This also means, only 16bit of 32bit are remaining.

And bruteforcing 16bit, that's easy.

There are just 65.000 possibilities left.

So I modified the proxy again, to instead of just using fully 32bit random values, to

keep 16bit fixed and bruteforce the other 16bit.

And in the packet parser we can then check the important bits to be all 0.

This means we found the valid input.

So I print WINNER WINNER and just kinda stop everything.

So I go in game, start the bruteforce and now we just have to wait.

I watched some TV show and came back to it like 1:30h later and to my surprise it was

done.

It had disconnected but it found a valid input.

So here it is, this is the valid input.

Let me login again, and inject the packet with this value.

And here we go.

Quest complete..

Blocky's Revenge.

And the door opened.

Let's have a look at that from the top perspecitve.

So door is closed.

And injecting the correct input packet and boom.

Evert output is 0.

Let's walk to the open chest and get the "Flag of the Block".

Oh!

And also the weapon Heap Spray!

Of course Heapspray is a Shotgun.

Haha And here is the final input.

How beautiful.

While my machine learning idea didn't work out how I was hoping, I think the solution

is still cool.

Through data analysis we found that half of all input buttons have to be fixed, and so

w reduced the brute-force search space dramatically to only 16bit.

2 bytes.

And that was easy to do.

For more infomation >> Failing at Machine Learning (Blocky part 2) - Pwn Adventure 3 - Duration: 14:34.

-------------------------------------------

Why Thanos Didn't Need An Infinity Stone To Defeat The Hulk - Duration: 3:16.

Avengers: Infinity War is a nonstop ride of block-rocking shockers, from the unexpected

deaths of almost everyone fans love to the crushing finale in which the villain clearly

wins, at least, until Avengers 4 comes out.

But one of the movie's most shocking moments comes within minutes of the start of the adventure,

and we're not just talking about Loki croaking.

We're speaking, of course, of the thorough beatdown that the Mad Titan Thanos administers

to the Avengers' Big Green Guy, who pops out of nowhere on the Asgardian ship to ambush

Thanos, with hopes of solving the problem the villain presents before it even starts.

What actually happens is that the Hulk gets beaten down and more or less given the full

"Puny God" treatment after his attack on the purple man with the really gross chin.

As a matter of fact, Hulk gets beat up so hard during the fight that for the first time

ever, he's actually left unwilling to fight on Bruce Banner's behalf.

But how did the Mad Titan manage to hold his own against the Hulk before acquiring enough

Infinity Stones to make him effectively unbeatable?

Before the movie begins, Thanos gets the purple Power Stone from Guardians of the Galaxy by

invading Xandar.

Is that what he uses to get one over on the Hulk?

The answer may surprise you.

It turns out, when Thanos fights the Hulk for the first time, he's not using the Power

Stone at all.

He's just that strong on his own.

According to the movie's commentary track with directors Joe and Anthony Russo and writers

Christopher Markus and Stephen McFeely, the filmmakers behind Infinity War made a conscious

decision to leave the power-increasing Infinity Stone out of the Thanos-Hulk fight.

For the sake of maximum intimidation, Thanos starts off the conflict at his base strength,

a classic Dragon Ball sort of move if there ever was one.

Joe Russo and Stephen McFeely have a little back-and-forth on the film's commentary track

regarding Thanos' formidable baseline power level.

McFeely raises the issue during the Asgardian ship scene, saying,

"Here you see the early establishment of just how fearsome Thanos is even without [the stones],

he's got one stone here, and I don't think he's even using it."

To which Russo replies,

"He's not using it.

This is to show that Thanos, the Genghis Khan of the universe, is unbeatable in one-on-one

battle.

He's conquered thousands of worlds.

He's a much more polished fighter than the Hulk, and he dismantles the Hulk fairly quickly."

The screenwriters previously discussed Thanos' awesome skill set in an interview with Collider,

saying they wanted "to announce Thanos as the biggest villain in the MCU."

"By defeating hulk relatively savagely and easily you're hopefully."

there's a sense of dread over the course of the rest of the movie for anyone who will come up against him.

Anyone who's seen Infinity War would likely attest to the fact that the early beatdown

Hulk takes from Thanos goes a long way toward proving the villain's threat level.

Not only does he have the strength of the Hulk, he also has a brilliantly ambitious

mind that runs circles around the Hulk's gray matter.

He's strong, which makes him threatening, but it's his brain that makes him terrifying.

But even with the filmmakers' comments taken into consideration, the answer as to why Thanos

proves to be more than an even match for the Hulk comes down to something simple: Because

it's extremely cool.

Make 19 movies' worth of a convincing comic book universe on screen, and you've earned

the right to occasionally throw in something awesomely unexpected like a Thanos who's way

stronger than anyone thought he'd be.

No one expected Space Grimace to be able to wipe the floor with Hulk like that.

Even on rewatch, it's a heck of a way to start a movie.

For more infomation >> Why Thanos Didn't Need An Infinity Stone To Defeat The Hulk - Duration: 3:16.

-------------------------------------------

ANDROID GAMES (JUSTICE VS EVIL - Play Woo) | LUV2SHARE FUN TIME 2018 - Duration: 16:06.

ANDROID GAMES (JUSTICE VS EVIL - Play Woo)

Hi! We are Fonny (Luv2Share) and Nick (Versus in Xperiment) from Indonesia. Nice to meet you!

Me and Nick are playing an android game which is titled JUSTICE VS EVIL. The best 2-player game collection ever!

If anyone is willing to add subtitle (CC) into this video <click link on this video description>, I do really appreciate it.

Hope you enjoyed the video. Thanks for watching!

See you again on the next video. LUV 2 SHARE. Let's play!

For more infomation >> ANDROID GAMES (JUSTICE VS EVIL - Play Woo) | LUV2SHARE FUN TIME 2018 - Duration: 16:06.

-------------------------------------------

Two game one video~KIMONO AND PET THE PUP - Duration: 17:57.

Hide the stories welcome

Hiya storytellers my name is drifterthorn and welcome to Kim Mona

Kimono something like that. You can see the name of it scheme I found off of

gamejolt angel or

It's thought I oh, I forget which one

But I thought it looked like an interesting game

Suffer you why not? Oh, so my computer crashed a couple days ago. So Oh,

Jack okay, so no volume to this game

Also, my computer crashed a couple days ago, so if my audio sounds weird and my FaceCam looks weird

It's just because I haven't got all my

settings back to where they were cuz I don't remember what

settings I have

Is there no volume to this game or is it just like

Really soft. I don't

Okay, okay, so this is a puzzle type game

Which o me and I thought I had that okay

This way

This way this oh, yes. Okay. I

Like puzzles games like this where you got to figure out

Every move, it's so cool

Hello, I guess a it's a little weird that theres no sound that's actually bugging me like I don't like

Silence it really friggin blows me. I always have to have like my radio on or

some type of sound going

Like even when I'm sleeping, I have to have sound I can't just have it be quiet

It's bugs the crap out of me. It's actually really buggin right now, like really friggin bugging me. Oh

Can you only

Do him when all the little guys are dead?

Yeah, yeah. Okay. I saw that one coming. Okay, okay

Interesting. Oh

Okay, so let's go back

This way then this way

Yeah, okay. Okay, I think I got I think I have this figure it out so go this way bop

bop

bop

bop

Shit okay saying I don't have this figured out

Okay, I keep on talking cuz the silence it's getting to me and I don't like the silence

It's actually bad and I would put music on but it would all be

copyrighted

So let's go then to this guy then to this guy

but and just

Pebble where I really like this game is canceled. Uh,

Like alright I could've just used some music though cuz

Seriously, I don't like quite games like this it like I'm sorry. That's all I'm talking about. But oh

Don't like ash she pops I don't like games like this

Gene okay. Okay. So I

Can go way out here. So did this one this one this?

One no, dang it

Oh

Okay, then this one then this one

Here

here

here

So close, okay

Okay, I figured this out. I love puzzle games like this

Like it kind of reminds me of

What's it super hot or something like that

I've never played that game, but I've watched like jacksepticeye play it and it kind of reminds me of that of like

Dang it. Oh like the freeze powers

And yeah that gives so the only way I can do it is by starting

plus what I like about these videos is

that

Like normally people in the in the comments like if there's a lot of people here like oh

Like puzzle games cuz you like, you know, you get yelled at if you go into the comments

They're like you should do this this one it was it was so easy

All you had to do is this and you messed it up so it's kind of fun to see

That made sense I

Don't know how long this game is. Whoops. I didn't mean to click on him twice. That's my bad. Oh

Oh shit. Oh

Dang it

So this one, nope, yeah, I knew as soon as I clicked it that I was gonna happen

If I'm here then here, but that guy's gonna give me yep

This one nope, dang it

No, the right guy's gonna get me but those okay if I go back here

Ah

Man um back this way

Dang it. Oh

Good - that way. Whoa, what did I do? That's a problem like with these games you gotta remember

each of your moves

Man oh I'm blended together guys

Guess what happened this weekend? I I'm so happy about this. My brother is gonna get married. I'm

so happy for him cuz I I really like Haley she is an

Awesome girl woman last night. Oh, I can zoom out. That's helpful

And I'm so happy for him and I can't wait to see them. I spend so long

I don't like that's the one thing I don't like about having a

Big family is like you don't get to see them a lot of the time

Like cuz with money and how busy people get a lot of time you don't

Guess I'm like I see them like once a year. Oh, that's it. Oh

Man, I want more

All right. Well

Anyways, my brothers getting married. I'm so happy for him and I can't wait to see them. This is really short

Should we play another game?

All right, that's plane. Oh, yeah. Look, that was a good game though. I recommend that game

I wish she's a little bit longer. I'd really like that a

style Oh

All right, so let's play hit the pump exit parties

There's no when I saw it had puppies, that's all I need to know. Whoa, okay

Oh

Wait and hear the fuck it's this way

Hi, excuse me scuse me, where's the dog? Where's the dog? Oh

Wait, I would saw that I saw that I

Know right is this game? What's a woman?

It's to migrate somewhere oh shit. Okay. Well, I'm gonna fill out this right away

Dargo, where are you Europe three years

That dumb puppy Oh Cosmo wait, where was that this one?

Is me scuse me this would totally be me at the party

Hey, there's puppies. There's dogs. I must I must pet them

I'm gonna closets. Can I go into it?

Why some good I saw the bed your clothes pop-pop you

Shit

It's all random eyes the help thing is this I

Suck at this game, where's where's the dog?

This way No, what the heck

I did

What

What's your name? What do you think milder?

X-files. Oh

Wait, I want to talk to the unit

Shoot shoot

Anyways

- I just want it and I saw that yesterday for my

Birthday, well, my birthday is Wednesday and the cool thing that my family doesn't we have this like week-long party

So what does he think Incredibles - it's a really good movie gizmo Oh

Knocking go find this dog

Is really good I like you too jack Jack the baby

I'm really glad you in it more than

People out of my way I must pet

That is a cat I'm disappointed that's a cat

How dare you put that in there Mama's me puppies

Trick house, how dare ya

Daisy oh, that's a good name of

Course, I'm going through closet right away point out going through. Oh

I see I see it this way

Dupa closet I need the door

Chloe sighs. Oh geez. Alright, what the Derby?

We're gonna get out of room what the heck cuz it simply he's coming out from behind me

See look out, dude

Got you get back. I love this game

Darling

Look out the other way

No desire to mess with me humans

Way my foot mining

always got a

Bandana around his neck

Know how to say that name

By the way

Hi hi, how are you? I like music when you come to their dogs to be found and petted me MARTA me okay?

Very nice hair you're smoking out of my face

Shit oh I got where are you?

Barking in

Gilbert

Watch it. Watch out. I am the best at fighting dogs. You're all in my way

Upstairs I even though there wasn't upstairs

Gilbert we are ahead Gilbert. I mean not that don't like seeing you

But alright different darkies

Oh

I

Like how I've gone from killing

All downstairs

Can't think of what a Japanese warriors called Oh

Anyways, I went oh my god, I went to killing people with my sword running. He's the best video ever

Let's get started

Oh

He personally made this, oh, I can't find this one guard. Thank you

Again

Rooms oh my god. Yeah the don't rat it. Don't move it

Are you serious

I'm coming. I have 15 seconds to fight but I may do it Oh

No, I'm so disappointed right there are night before you

Pettit it puppies. Hey you explored 142 rooms

You talked to Stacie want people spoke? I think you went home and read Campos. It's Queen fell asleep. Yeah

This is a cat

Not a puppy

Starks

Anyway, so resting this here eat if you guys aren't gonna find all these bodies you can that's a fun game as well

But thank you all for watching and I'll talk to you warriors later

For more infomation >> Two game one video~KIMONO AND PET THE PUP - Duration: 17:57.

-------------------------------------------

Book-Tube-A-Thon: Book Cover Fashion (cc) - Duration: 1:20.

Hi guys I'm Shelly, and welcome to day......

Whatcha macallit, thingamajig. Yeah, I don't know? How many days are we into this thing now?

5 maybe? is anybody? I've lost count.

I'm reading 7 books in 7 days. Trying to do 7 challenges and trying not to have a meltdown.

It's hot!

Today's challenge was to dress up as the aesthetic of a book or a book character?

To be honest I was confused. It was basically to cosplay.

So I've decided to Cosplay myself as Cherry from All That She Can See by Carrie Hope Fletcher.

Cherry is a baker, who lives in pajamas. So the pajamas are on! She wears a headscarf, she bakes stuff.

My Niece and Nephew may be coming into town, and I may baked things.

So that I can win their love ,and they give me hugs.

I'm not shameless at all!!!

Here is me, in all my Cherry glory.

(Brown Eyed Girl by Van Morrison)

The radio came on, things happen.

So that is day thingamajig of the booktubeathon.

I hope you guys are good, and are having fun with all the reading.

I will see you guys tomorrow.... Maybe? Possibly?

I don't know, my Niece and Nephew are in town, priorities people! Byeeeee

Can I stop dancing now I'm knackered.

Broken people really shouldn't dance!

Không có nhận xét nào:

Đăng nhận xét