Monthly Archives: June 2017

If AI and robots are going to take our jobs, at least they can do it in the most relaxing way possible. Just like Shimon here a four-armed marimba-playing robot designed by George Techs music technology center. Sure, Shimon is just the tip of the iceberg when it comes to AI making music, but just listen to those jazz-fusion vibes.

Like many AI music experiments, the music Shimon is playing is generated by a method called deep learning. This essentially means mining a large amount of information (in this case, a dataset of some 5,000 songs) and looking for common patterns in the music. For example, if you have a sequence of notes F, G, A, what note will follow next? Deep learning will give you a good answer.

Shimon the robot has been around for a while now, playing alongside human musicians using pre-programmed songs. But now, its being used to play original compositions. The video above shows the first melody Shimon ever created, while the one below is melody number two a slightly faster number:

As weve seen with previous experiments, the actual musical output is a bit avante garde. Researchers working in this field say this is because the deep learning systems we use to analyze music tend not to be so good at thinking about long-term structure. They analyze the music in short bursts, and the resulting melodies sound quite abstract. It is possible to program in artificial constraints so the programs will produce songs with traditional verse-chorus structures, but at that point its not really AI-created music, but AI-human collaboration.

So, although the marimba is a particularly non-threatening instrument, the melody itself is proof that machines have a way to go. For more information on Shimon, check out this interview with its teachers, Gil Weinberg and Mason Bretan, over at IEEE Spectrum.

Read this article:

This AI-powered robot can play the marimba, the least threatening of instruments - The Verge

The maximum score possible.

It's official. Humans are now second-best at playing Ms. Pac-Man, a 1980s twist on the arcade classic, involving eating pellets and being chased by ghosts. It was rated as one of the hardest games for an AI to beat, but that didn't stop one.

An AI from Microsoft's Maluuba team -- a Canadian deep learning startup the company acquired earlier this year --has now scored the maximum score possible of 999,990 in the Atari game, beating the human record by four times.

This was achieved using a method of reinforcement learning called Hybrid Reward Architecture. The team taught 150 AI agents to work together in parallel to master the game. Agents were rewarded for eating pellets while other agents avoided ghosts. A top agent then got feedback from the others and used a weighted average to make decisions.

It's not just Microsoft's AI that's in the business of beating humans at games. Google's champion game-playing artificial intelligence program called AlphaGo defeated one of the top Go (an ancient Chinese board game) masters last year, Lee Se-dol. This year, it defeated current Go champion Ke Jie.

Microsoft's AI's success at an unpredictable game could see researchers help AI tackle equally unpredictable real-world situations.

Special Reports: CNET's in-depth features in one place.

Technically Literate: Original works of short fiction with unique perspectives on tech, exclusively on CNET.

Read more here:

Microsoft AI gets maximum score possible on Ms. Pac-Man - CNET

The race for artificial intelligence technology is on, and while tech giants like Google and Facebook snap up top talent to build out their own AI-powered products, a new startup has just raised a huge round of funding to help the rest of us.

Element AI a Montreal-based platform and incubator that wants to be the go-to place for any and all companies (big or small) that are building or want to include AI solutions in their businesses, but lack the talent and other resources to get started is announcing a mammoth Series A round of $102 million. It plans to use the funding for hiring talent, for business development, and also, to put some money where its mouth is, by selectively investing in some of the solutions that will be built within its doors.

Our goal remains to lower the barrier for entry for commercial applications in AI, said Jean-Franois Gagn, the CEO of Element AI, in an interview. Everyone wants to have these capabilities, its hard for most companies to pull it off because of the lack of talent or access to AI technology. That is the opportunity. The company currently has 105 employees and the plan is to ramp that up to 250 in the next couple of months, he said.

The round was led by the prolific investorData Collective, with participation from a wide range of key financial and strategic backers. They include Fidelity Investments Canada, Koreas Hanwha, Intel Capital, Microsoft Ventures, National Bank of Canada, NVIDIA, Real Ventures, and several of the worlds largest sovereign wealth funds.

This large Series A has been swift: it comes only six months after Element AI announced a seed round from Microsoft Ventures (of an undisclosed amount), and only eight months after the company launched.

Weve asked Gagn and Element AIs investors, but no one is disclosing the valuation.However, what we do know is that the startup already has several companies signed up as customers and working on paid projects; and it has hundreds of potential companies on its list for more work.

As weve been engaging with corporates and startups [to be in our incubator] we have realized that being engaged in both at the same time is not easy, Gagn said. Weve started to put together a business network, including taking positions in startups to help them by investing capital, resources, providing them with technology and bringing them all the tools they need to accelerate the development of their apps and help them connect with large corporates who are their customers. The aim is to back up to 50 startups in the field, he said.

The strategic investors also fit into different parts of Element AIs business funnel. Some like Nvidia are working as partners for business in its case, using its deep learning platform, according to Jeff Herbst, VP of business development for NVIDIA. Element AI will benefit by continuing to leverage NVIDIAs high performance GPUs and software at large scale to solve some of the worlds most challenging issues, he said in a statement. Others, likeHanwha, are coming in as customer-investors, there to take advantage of some of the smarts.

AI in its early days may have been the domain of tech companies like Google, Apple and IBM when it came to needing and commercializing it, but these days, the wide range of solutions that can be thought of as AI-based, and applications for it, can touch any and all aspects of a business, from back-office functions and customer-facing systems, through to cybersecurity and financial transactions, to manufacturing, logistics and transportation, and robotics.

But the big issue has been that up to now, the most innovative startups in these areas are getting snapped up by the large tech giants (sometimes directly from the universities where they form, sometimes a bit later).

Then consider those that are independent and arent getting acquired (yet). There still remains a gap for most companies between what skills are out on the market to be used, and what would be the most useful takeaway for their own businesses.

In other words, many considering how to use AI in their businesses are effectively starting from scratch. Longer term, that disparity between the AI haves and have-nots could prove to be disastrous for the idea of democratising intellectual power and all the spoils that come with it.

There is not a lot left in the middle, Data Collectives Matt Ocko said in an interview. The issue with corporations, governments and others trapped in that no mans land of AI have-nots is that their rivals with superior AI-powered decision making and signal processing will dominate global markets.

The idea of building an AI incubator or safe space where companies that might even sometimes compete against each other, are now sitting alongside each other talking to the same engineers to build their new products, may be an industry first.

But the basic model is not: Element AI is tackling this problem essentially by leaning on trends in outsourcing: systems integrators, business process outsourcers, and others have built multi-billion dollar businesses by providing consultancy or even fully taking the reins on projects that businesses do not consider their core competency.

The same is happening here.Element AI says that initial products that can be picked up there include predictive modeling, forecasting models for small data sets, conversational AI and natural language processing, image recognition and automatic tagging of attributes based on images, aggregation techniques based on machine learning, reinforcement learning for physics-based motion control, compression of time-series data, statistical machine learning algorithms, voice recognition, recommendation systems, fluid simulation, consumer engagement optimization and computational advertising.

I asked, and I was told multiple times, that essentially colocating their R&D next to other first, for now, is not posing a problem for the companies who are getting involved. If anything, for those who understand the big-data aspect of AI intelligence, they can see that the benefit for one will indirectly benefit the rest, and speed everything up.

That model is what made Yoshua Bengio the godfather of machine learning so excited about co-founding this company, Ocko said. That massive research advantage leads Element AI to be able to deliver technically advantaged, increasingly cost effective solutions. It means they dont have to treat AI decision making capability as a scare resource, wielded like a club on everyone else.

View post:

Element AI, a platform for companies to build AI solutions, raises $102M - TechCrunch

AI isnt going to be a threat to humanity because its evil or cruel, AI will be a threat to humanity because we havent properly explained what it is we want it to do. Consider the classic paperclip maximizer thought experiment, in which an all-powerful AI is told, simply, make paperclips. The AI, not constrained by any human morality or reason, does so, eventually transforming all resources on Earth into paperclips, and wiping out our species in the process. As with any relationship, when talking to our computers, communication is key.

Thats why a new piece of research published yesterday by Googles DeepMind and the Elon Musk-funded OpenAI institute is so interesting. It offers a simple way for humans to give feedback to AI systems crucially, without the instructor needing to know anything about programming or artificial intelligence.

The method is a variation of whats known as reinforcement learning or RL. With RL systems, a computer learns by trial-and-error, repeating the same task over and over, while programmers direct its actions by setting certain reward criteria. For example, if you want a computer to learn how to play Atari games (something DeepMind has done in the past) you might make the games point system the reward criteria. Over time, the algorithm will learn to play in a way that best accrues points, often leading to super-human performance.

What DeepMind and OpenAIs researchers have done is replace this predefined reward criteria with a much simpler feedback system. Humans are shown an AI performing two versions of the same task and simply tell it which is better. This happens again and again, and eventually the systems learns what is expected of it. Think of it like getting an eye test, when youre looking through different lenses, and being asked over and over: better... or worse? Heres what that looks like when teaching a computer to play the classic Atari game Q*bert:

This method of feedback is surprisingly effective, and researchers were able to use it to train an AI to play a number of Atari video games, as well perform simulated robot tasks (like picking telling an arm to pick up a ball). This better / worse reward function could even be used to program trickier behavior, like teaching a very basic virtual robot how to backflip. Thats how we get to the GIF at the top of the page. The behavior you see has been created by watching the Hopper bot jump up and down, and telling it well done when it gets a bit closer to doing a backflip. Over time, it learns how.

Of course, no one is suggesting this method is a cure-all for teaching AI. There are a number of big downsides and limitations in using this sort of feedback. The first being that although it doesnt take much skill on behalf of the human operator, it does take time. For example, in teaching the Hopper bot to backflip, a human was asked to judge its behavior some 900 times a process that took about an hour. The bot itself had to work through 70 hours of simulated training time, which was sped up artificially.

For some simple tasks, says Oxford Robotics researcher Markus Wulfmeier (who was not involved in this research), it would be quicker for a programmer to simply define what it is they wanted. But, says Wulfmeier, its increasingly important to render human supervision more effective for AI systems, and this paper represents a small step in the right direction.

DeepMind and OpenAI say pretty much the same its a small step, but a promising one, and in the future, theyre looking to apply it to more and more complex scenarios. Speaking to The Verge over email, DeepMind researcher Jan Leike said: The setup described in [our paper] already scales from robotic simulations to more complex Atari games, which suggests that the system will scale further. Leike suggests the next step is to test it in more varied 3D environments. You can read the full paper describing the work here.

See the article here:

This backflipping noodle has a lot to teach us about AI safety - The Verge

Facebook is teaching chat bots a new skill.

One day, the art of the deal might just involve lettingartificial intelligencedo your dirty work for you.

Researchers from Facebook Artificial Intelligence Research (FAIR) have created AI models, or what they call dialog agents, that can negotiate, according to a blog post Wednesday. They're publishing open-source code as well as research on those dialog agents, the result of about six months' work on the project.

The idea is that negotiation is a basic part of life whether you're picking a restaurant with friends or deciding on a movie to watch. But current chat bots aren't capable of much complexity. Their state of the art is to do simple tasks like book a restaurant or have short conversations of limited scope.

FAIR worked on the problem of how to get dialog agents to operate like people -- that is, come into a situation with different goals and eventually reach a compromise.

The effort is part of a broader push by Facebook to get us to use chat bots. At its developer conference in 2016, founder and CEO Mark Zuckerberg walked through scenarios in which you might use a bot to interact with a business, for example, to order a product or get customer service help. While tech giants like Facebook, Google and Apple are keen to build the personal digital assistant of the future, today's helpers still lack the necessary skills.

It's just one stitch in the larger fabric of work by Silicon Valley, academic researchers and the business community in the area of artificial intelligence, driven by powerful chips, fast networks and access to massive amounts of data about how people lead their digital lives. That's showing up in everything from sorting photos on Facebook to beating Go champions and diagnosing medical conditions.

FAIR didn't delve too far into what applications might be appropriate for bot-bargaining or whether this capability will surface in any Facebook products. But the post did mention this could be an advantage for bot developers working on chat bots with the ability to "reason, converse and negotiate, all key steps toward building a personalized digital assistant."

Negotiation, the FAIR post explains, is both a linguistic and reasoning problem. In other words, you've got to know what you want several steps down the road and be able to communicate it.

In one example, dialog agents were tasked with dividing up a collection of items like five books, three hats and two balls. Each agent had different priorities and each item carried a different value for each agent. The AIs were taught, in a sense, that walking away from the negotiation wasn't an option.

The ability to think ahead is crucial and with the introduction of something called dialog rollouts, which simulate future conversations, the bots were able to do so.

Or as FAIR scientist Mike Lewis put it: "If I say this, you might say that, and then I'll say something else." Lewis said those rollouts are the key innovation in this project.

The research has boosted performance in using various negotiation tactics, like being able to negotiate until there's a successful outcome, propose more final deals and produce novel sentences. The agents even started pretending to be interested in an item so they could later concede it as if it were a compromise.

Humans had a chance to try out the agents, and the researchers said the people couldn't tell they were chatting with bots.

CNET Magazine: Check out a sample of the stories in CNET's newsstand edition.

Batteries Not Included: The CNET team reminds us why tech is cool.

See the article here:

Outfoxed by a bot? Facebook is teaching AI to negotiate - CNET

nZsy|U tQX/7%ip2cIRN_+0oi%X8/LA1miO/(vyVeQ}{9^yz3"J4#KWBRPstW@x h(86EEC)Z&IZHNxj[j5C9qB~tnYg=%iiEaz/QPVa.)!Hj4irT2s|'~nXr^ilvj)3G ec; qwX 9/$]ig0p^T=gz4E9*-&uZdZF";e2:(Q9@62aW]MCgtN; _vQ@rS />qV/ad:IOzGvo~/eyx_} !S~zsoHg:=tStl=Y`Sm;PbV{V0D[G&a;q %1>>`=mSZYT{ac7 olCB?nE [ Juyv9.kD4uLJz]{q0'CrGiv`6 wFTv@/8v)~bhAzt 4QQR&=MtlobqdX?+cVfy:wFKdU3;>z >ZS13g#`:ggj"1NR^nw]pC0pb? :hjTR35zQkx) &x&'l%EchbW>1'We#/rHp5q,ifhZSC?S1]TFLfdrOH/%mN]{K07Ea#VW{ZWXN[u 921f{7_ehhU,^U?.0_i.&!Zx?!kd3H{??]D96[#k0t|'h2l?BA,u,v*Gk2y4_^`zW2? Z -g)lsOPb8w6>?-V7-py#GhT9r9k?sC.;ZUUI*v4c3 %57O`}o 4y,h4-:b_G!L|{!GTe q>UR0BSXB^X6$FfPUhLrBuh2?DK-qS+MQAC#t#gYDXuy5ayw->h?E,@ )@%RBouF=Q O!ys}TPx/B}|n8ah4J5eEykpx;lB/)K[=qV3>t#m$i7B{Q( Mf~(3^7Z&3^gHm>g!^SQv(59.'Jyw9QPt-T=ml2uIJ 4|Of|Y V9?UE~Dd+x]w5e=^^)I?e/qzt)Ia7{Wy?3wu cd!|{&pP8!'={919: v)Vn>TDX"$')A-c.}V9a`(N0 {~_>wYRvp6Ce^2+u~Opi4yCgB$w;%G`6TexIQVmh -4|YsR~$ 3l|{| kxietsms}n9"q%Cx]{mX.+uVBR+`I|y [0#Pov!#w{>7(KS sX83P@*(nYpM][Y?@@y4/=KUx757 D$})tmS[DA0sL' }I7[Jm9O-{|p(1A5xX-#^,2*zH j;4g4QKgwO$e3WmFwZsnw/ 0+>4(Lyw*L1Et&7,G r~eneX0=C:tBwtM@lroc/oC3^zZ ZSP>JmK>jx..RKi*Pal r>FA (%T,dh.rr"IsxQ8u,2[nh Y7|_n %*[f,zV2]w-5M6dusmAcOe I{!E'aZaIv_,y4']Fl.l$>_]LTtRu|7ANlC8@ sg0Nf*(8+d;2Nif,e ,.ap$:#wG2ZRq_R=z3*c,45f/s.Du>q}Yo'-/?|`)iiO4 =aWA=k^z~^wRR?; ~F:^b*h ~p>oee{e+2~ 7Ot#0erk7$G1OqGCz' d%4}Mv'xby>nyU54Lf+p8`MUK!L73M*B$(Kx{u,,FSm#c| 5lbcyOTMTWb"hCyf W#21`VzV9 -9Iz BG>IDL/q=(=}%4{bz$div}84'@iFT=dGY:+];Xs -B.k-/jmEeWe4HN($!/ q/R++ @r2s :%~foXrZ2 _MTI"9=Q2{:>;2CX;JL>T6z^}03pFiM6w*|NuWS]7#nhX"2duX&n:nm+ =TE9e]^t{eA5=|RIA_r!+ e^J)Kf2QG_el]x?pwFg31CNUs c{Nekv" 8!q' r=Dwfev8X'_/_-w|s~!x5%^EM?Sb/2}}%y0Kbgz%/>bpxA1k*1ou8O[Y5_">P@8o:3Ww0VHA-xR_j0@@aql:nn{}aO~f00~Zl.fO]GY{1-+42&C6[L8GoPKKjVrbjzfzs@3LLH2=r]G2M^]2FFKq$^C-[03g7a|`G{*0eF@iI?GS6z;j%>*PS ?WFX`-Jjus?94xx (e$]w/G5Tf C0r==W.qz N28_Dd Fe=s$Y5?b2}EZXjp~} z!9S>f@q_x=(Y!Lb4':[r,~!2}MtW^wq$)EXWax~*xn?`[-g97Ytn_Br;I10fVx> ve$nhyY3nAn :#M5c*'[9LAWib3|>tf,Aq|- )Y"p1fN.9 H506Rb?E#xXFl$D0{+s%i'y@6X8nY-/zc .&"R%T,EBH&7 X4?rShP|yo@,o|Y04G:YT,CzZsY[ym|IVUKG&A{J}KOK4K &,ab72/$JHLLW6SY2( 1P&QmHd0TA>@' oWxxr0)xQ81m,+X{7-0n=~SXa]P.5'p5tQ`UR.XX>d *#M .NM 'pMjq@Sf {JiO8ms9y=`H..D[MK!V)t 2Brrqgy)M:X$0v8' _+o!.&Wb>WLmeN9L5WhHt@{+jE2U&k(Y>oxx'kbD}X!k^&`N"6p5ZcEIQzsYzg-j=]/XD>gBtn3:w{!^9f z~_bw7WF?2?w8`839A0Z9x}EF9u-8~QkW?7{;dl+`2 D~-V5$S))}b!db@ CT %;Y1B`#1 W/(@>(W-r[{NP.]rJ;V>c^GIM!;3:DtRXvpRrHV,j9wBZW2jV$pEkOC/~Pf/17F&tk3w"F*eo2MW>`AO%c(X8Oxe00[./o1bS4*Z>'y*RXaODiz!;tTp2(C4hBr~-i9T2x $@flB?Qt7zxo:C.QOO.ep.Z$_ _BHE eg)|[]hk1%XcF1s%dIW#.XL)-! jfQ)PQg6E {fj '2R*=!xm k B%N,xb[DA*5%x@zRoS3 gGe]kb%AGy:x#( [KCbe`_!WpF:mQ1*F#94@(P3^~3ASd KCR.qD@k/ERmHl;'~`=eyf/[[K/>H_t/Bh`"cd $,gP,QmDZuKSpgP9*uJIbgi }NCJA?Dya,J~Z-B8t1^@D/[ZQTGyK%q[Bt 48PZG/23?H^:Oad;d'8]^bl51`[P$*`gix'5^,8MbX~da$bN0]6F|L#zix,:Ui"xTXBp.eQ / ee'He8h bjR/)aq!2HjD%Q`S]8hyL:vY,Sbxd pJc5Ny_d%BLEAIgQHLpY+"n!SZL{/QI5S=n>IE3sp# h:rRP2X6.=(9Hq(N`)&J02*>SP0vI^ Yupi $,G|Ka QB6F5i}R1=BdwJ:#J ) )] Hb"FRa0$jjPAhnda+1af/.o^ru2gzIR2DK>'_uz sxU3dB.[=^VHqR2 7)S saU^E$.z6#r7%w'58=VML$'d+q!/VBup90r}gKq&nqg`g/;FjZt7 }^oIE2]P(4KIW}D&G' k_J!oU2atgZReM0U"3c&IBx)2m%@PUx+NKLRkokx7u"[BJ^PI#B1MzDRiPK6*1WQKe' ;/3uw NOd$a&6ZUMV/I 2B*u@(9aV'PQ@&:X }KAiI[fB*ce$cx[}ZhK2%JV!7N$|0n@p7r#{qoL*0O`&^ +e{S!.USA42 $yg`x,j M9sk-6R)gx$%Ul?1LGBKs*!1+@D>g (y>jhw#I H-X=pra+e+;khV'.Bg_4[cUHS>;*6atV]L[ku3M->J ~l4-Nn' $H +]B8Bj/`nCr~&Go P'm+|n)9$d$F& hft702`)8a6 Gg@e2lGU,9{tCuG u.bvJ}7k97tQ7;3HV;1l'Q*9D|4S3T2zh2Mg=6s-WSyO[u6Mx'TiVc/,Yz\`y'N]xzcfdTil ]nl*si_ C=tV,,^zrT"=A&F&mU5}MZRbQ>R!GG$u4t>{ h?5)&mSIN!B]wKYAxdYVi~ #h8jhk.Gagncs**_Dq^"XoGe;:/ M;rz%]=ML2/M#e*WFLfdL Ho&irQ%a=Dq G0)3(Bq$q7"(*j&J+* 5iE=tft*K{WJ-;pidiZ/[BiXVGYX{>c&$e [ jPx4@V*R ^cNVC5Dx`r4B@(;%4P8YQ`p~.)@-Lu8(O`4`xwU3&acDT gi&bF`c:#" (k,g(&3v?H$*t1rxA8mGg &=?wJl[g)SKDw*!*flN&n.)yd?8Kzw1SQ+ 22 HUg6v|6]r4O, ?5MvzS,7%2v).!|S] Qaf}F~[P94dqY,l%L~Mukqgjhq4sT%.QqBdID0_!@k$}nA vM&J^RN0X;%RH{QuDH"q?-D7Mot_ ZTQku9S,g2S0z+S}]h=bU&@dxIp)! '*+Koh 7*GF!2 Y5qKJhA7uK} YElJGIFzE8OX8tj-;pHM }Her*LXu"!L7CMJ*fSuW!gxBTJ B*e![/60|7d >QC2%-Hde,U%!`jQXKDz|Za.Xg6Q"FD="PI#Y.+(^1ZV w3~fH4H7$ L]vI^VnH+CLBD$Om!%pPKV^K-kJV @:T2&s^0MF'oie+>7+FFholHK+ule )V~0%rw^eiMh49$Q@KU?Zu=_yA- + ouZRU-dT_oi{?4d|;rB;m -n E5q05(,=/[%Kr`3tD-ni)2W4[iAQb XqpX,'C)6enU+kDs_*;ei3Q,jSU3O)nZ^sy`J!LNKE };M(255cR8UZE~GY](eV4xO&48eaLA LoAfJ$AdM,ZPZ.rnvu7&kHnRu2M l,bWsonW0Y[|veAbw_.7j}z}77qtj8 9c7]X8#KOD>[/DtwwuV}kFP;Yo(t`s[boQ9|(:_VsA*HG6#0hx7ORKOTmt*}fFkZWbWv>IY0w,fuqBPqLI]/}JX=V+i?{ef*6WB_TVu WBem)h(r2^U?>z+"P3&vU6FSg4.UvugO??w-Y_6}?kmC0{fo(>wa4 +]L& kc~R RuAJf[7*wi=>}FLK,#W@~rS?*oxq/(;RpJv{=Qbb[g)x iH{|=>,KxSi6t5o*}Y*:qAI[B"Q9?S^SsCH)6X{Oquj*5N(=>b!'b(t29~fM64Kh`~U

w{su~7.~l0:?&.~3|stsu~wN+KUVqL;&f4K:HD7_Jahb'Al1vjx ! #h$~8*Nz^7?/,z^zpju%K^_=+-Q4:Cv/4geuxb B^0^ *MEM UuOu?|%y?&b'H8y)P"2$$g*xuEb%iSVUf9)OJ.-:zk/$J)F"^AGSz $*fvbR9R6ag) U. tG#BUHC"3'H#|#Li;+*vi"V*"?#Eh1cy?w:ShD|ZqY[CAx{h_~ {.G J|q |aZFGI*oBPrQv?!FDIyx=4fal"E,)P:Z@fBiaIXxf:[HIsxhhlR"E#|H]FY+-Rsfl]>wchEV2(V15(-u~ent8y d?.>=zj&{'y8LfICb`> VWzd57s*Rbz$BrJ|2+ I+ aK/,9W9b[ ;n0fQoh fU:hc>?{.Y-,;[ :ceAr RL8mE6bF:$,Pa^^d:MZ' /M>BG!xL{,zV/VW> ?/yA;Q05,+=U?^J#tBDk5vZ| x]LudqMHN[yI*-h&x p?YumMJcGepf{f*a*yi>{EUc,7!lv5(UsZM;^qM)RXo{wlzt0:.&wKNtXT_pu{mg`FuymOjc?U3N7[QOLe+?)j +G2&cgJU'RyNQy$|Rv S$o7!2"#9Uij5*} bk(JgNzhq-z(th^&7RPd} !EXN?p8KY7(u3l[v`YX$c`'*XOggPPKHwL%[}gL#~~;r ^U_fzY0LG7qkYV7f*5 ]o-]d[sz'D=cy/06AFWY)t9ek"P.ih+?'&Lu% g)".v>s}]^`?$.U,)L4JrE`3{|5o?J`&n?x3v%E,k>1Ytyv=ru?6';fymHj{!L&Xw No-0/!To n10~5[^bqwINWfuj8h~WTvWv6 PiZ ^MDg24]rZ &5qj>sO::Vm|nhGI0!&;;{NY% -ZLp9Y;c4&g p ta'!b4S3 e$Y,cOyS) SA[3V *2{67Ts9_m[OBNsD@8OyFz7?x*xOqI,Era76T JLL4#| xy~v%SVzI=mw MutuMl2g]s&h`&k]'SAh}plLnPWFdd*iXz3l3W/N^7u-P8O1N`pKWSqHJOUgcl^8{9Okq.O'{*eSs{v^tSm0b++{&jyI?y6vtN5 t.NKPp|q 4(6B>u!=+;A8Z)pdGG+yS!n0op.3ep(yU3w8;l;cW9NCh?`LM.^|c[Q"PU} }5?z~~rl8co0O2WR@3x f,,]'pt/A9O^C>KP>Sy ^=`cnKP6{f;KP2+NrV/Ap8}/AQ1P1L}`cnP:f;P6`EBrT x:=dzxwz_[]w^z *MK+3?_C+zs`0lbl(rl&a?ZZJ 9Ro|'6irRw&k1)s7E'rB XmNcHm{mAg,o1x+0Zqw>w;/zl&zVi9V+f9Z}M dGyZ]:vjiveVsoL.J@v:w.eU~]nzY0q,i-&;Fd@tjmw|f,A; xzj!'l#QVD&ya&tc Rk~t;A>~GiW/(uAPbSW1qE}0[[@]e)ta X|S-jTA5voY{:5Yp;[8YW|+C^i^uuGUN7J>zN8Ozrxw{v&,wEW"Mpj0'9unba+,cMng66ZrZC!%AY1-E9ym&_pvw``OP;^NGxv_]?DGZ:YrwgaTE:b&lIT8|_iAr;V:RSll. RR 3R C^&KGhHC6D1>7ph75Kvjo8t7WO?Tmq@o2@o777}{lU%,konD$YH[t&ZuGqTNx`p g%_4`)gL MJ[.Si*f"Xx)NYsF0G-X$UQr"_zx/E^dmPqp,2I um2|aN2d,@sgc>,X/=!|67GoCglj9.RdV+qJD'b`Q(ylyoaGIRntc"&r@s="%#E ~=711VMLG@&@m/KNo)HQ>3-NE$ Kq+T2y;$X&qwTOrC4@*vV`PbL%jyy.=l> ~>

3IY#,'xF8Zl-?V9pX~d*jR@32j9[Dz:dC#LBKZuhJnxTaE @A"#E.9B#X"m#W JEH{t$1Fr*3(HP=-*DoY712$RIdl,dYs(J~^~Y e>@6Vix8M/nQ|K WUbmX/CU =WN?M0kNDBA{KCTAZbclMu_yKC}Q77a*(HB%%xb) H|.RCw Y.K*fC,#d%C(>8p 3~"yt1y:IR$-x8G|pp]H?Azr+vG^#sHp esMz6Y8Bad q T+MZ!=d[Y=p0i-9{-1_dhn%5+Fr)) 1yY, 2pdlT) =}GzxtHd8RAI"C:kTS IrT&c3 SA lP(R7/sf~jTJMR gmo 0}Qg!3gm|2'pbNLae/$z}' 8z9ARZIPDpWP5jdx ~.d|T(TW fi08 `bAsdL|:Qw?X-}33I|crZg|jMf >tO%vMCltiG5.3$73!~WC f0sn|2g$gq,HZ[D7i{-zO~cmS=|=f}+R,dvb(TV`.TGNU *PB8y1a`,8n?u0>L 5U)WTNjFG2M @g5Kc{rKL.?%0p~!obf|>|t@a6>v%>{?ex}FBwn"9y.kZh~^Q 6Pk#lRb[Thi swl3gC9:t0uC:>t8v(`fSlCN{0?j:rrzmj';mk7NrW6+ym^X wapE7j3[L9Pw@~I!.DA6B*J|=0s7~Ja_N`5~HEpbN7=6xC$Pc]{H??Vk+h4J7.d8 jl(uC ZMB=r.c,>a+Xzz${'Si,V>UYh'c=HD&5N#G$Q]Kj4qwVPa7&Z& #v:}Gjq0B#OZa6*p`_#Mq0Mu(&])sBpT(?YR R,,bdj.$_k5Or *so}y ?v_P4L/?1F{KW}Bqpg{PExN/EhtyW 7r{OBZ9tZMVi11swj7G1 wRg-22M*xwk'IQ4NIYkMfs5)f9InM;jGyIY>v&imAI9hfRo`RfpGJ:-F#L@FFN&@L-o0)'~[K!!x~;/~[pZ{0glb(+UKq@ 5s`fvxs=?J"_v(,~{&"`kfdhN{Jjz~l 16dxgN3ANdCw?;z7UOvhZ49yxki;Ucg&nr%Iw? h=@G/voTW1h{XjhHAP ?#t|#4|9P[$9_MC'C& }O!`tZp z5etQ_VJsI}S n+D$_b^5ZY,^G/F'opNmn|nOiLea68Dr5}gVR1Xs):|Luyn5YBG=/TLc,~{#50JNh$Hn7*;])MPJ'lrF).>+6+z:3~,3d/+Nz, }&'3v[6}+5+p:rg@3D'(/NUKg:4aLg=/9_[WgsGY-OZb3YhTqnsr*cAJA=L#ybv[N-}lk2&c>fc{nWfV}xL|?pfpibqc+AQCx07c'o~l/x+S}al$8hIlWkIqW /0`.$iJg&)zS0I-l&m`.'15Zj2G!dvJ2a^M[]LJr_r{WI3WU *mb&b@|w^@6LgyN&Jwcg SgN{>Y}/0OID*S$gb sNj*B? ay"73zk"JdpjW-H@Ek}hw@1$"zkC?KC7 }eu6Q0`Grjn5?Vf&zm(;W}0kWz+5OhG4sH Xd>~*U[P+xB[ m8xkv tZ%dUIb` szbdi+ns$pf{f}3QD3A;x_uU{?uSw4.wt2ia:mqjpt2k'>+2HqLR_A70q>Nd?: ;['@woZGf1{S7 RO%s_(c:X?S+'cOOPHUVxC_~_h_K)IN[>p4IW8Zq(o>Bkx xQ,1E 5iph0 {Mqvmq): M1fIU_q,:)5XWDM- xvuLRRln_Qs$`q_9JP%H|BBls%Ek#~A,krQl+ESQfs`0W.:Sy@ok6'p367h Zs1"M>:"P>kE~mc""P6"f;6"P2|C%Y!SEjDxgI2"PZLSsxTF-} n+q#~-B* )DO E0>MfGu98ppv[;Q"Ad?^v}4'8, bEksua/C#+jyd48@dh38@8@cs8V;-s MY#>F zn9Eaifx> qf/[SOpJu /?7Lp6 29t2bVYdLk0w~S|8Jl `j9I!;-n5Y]G{,YU7K>J4W;FH;CNSU,[EA]up`63]IlHS8 @Gf) =x(ML2&zHK+|5H=[o&}4n|V3s;03N6?>3(7m_)xpq1y`72z'r[gaNWNi:%m$ -mzh4s*q0Eg'$k:|cUa ]zthb,v[Mxh_(qKUW TCWWQi%Qms6:dlumA''flb(8 _Ls7g5lc|(51F/D=JDl0l4/|4FDTDFF6CB~WQa;3IIzV&:A:SM*[n^>AVz&Z969-BkgYY-fNi :mu[z}otY35{N S:K@VASJ *^3xrMupB`b'-(|g}>lhk0!__rKMn/aDb w8.p( W']%p?6HW}`}+]VdK%,-'*}5k7N?XL|Fsl o LI(%syxQG}&OJ*KwHJE'&(NS8nVER{Sr?GC_Smp>H_"o?J?U.A3?aw@cQWPo $!wWj8vvnu@kl>U& I[%8=x1r]J;~})R$nFUkW`%pokJt*ar(op $xw]g=K>lHQ-7gk}"@0w *'naX-?n!Zo ?voik LM lHsOT:--@Wd ~:7}=g*X}m. e3%x3@8Ol;&T)(ERMoxMa~qw)5_5_m`s g5f*jRg]5MBgg4LP@~U=yRW{9ym_eVG__o6!c+=o2,t 6;cR8f j^ *is !h}Oz.G W4t'y!fLvaT;8P4L=G]D@G?kN~^ "3,Ohpt4Fl1HZdNGG|0u}4k=8|# |h{v1J!g$r>8hg!sL"l9ev J'(]!>}mFSPI$v,v;=T8djd8T !Hh_ZYmloO{tp|mAL* 0E-.%1,yv6}o g>@^@l_j]'L6sgMNOT_[sb3Y;/%M.I2 (D>8a]Bw+!8K|!Eg`(Yv"W,E>vT& %.uBd`[^JdIn9IZKL])UL&pb7)Fx,u9BBU=geVaD3/ DI:"TJ(^#txpQB&*$k|Ho;TPv4"#u;D^"b)}7NH>CIF $m.[p)S }`._$n7HwI2)&M$5$cx$|pKe)Z]/5CB^l,2T0J,Gp I%q(l[XP@H?Re`#qc%%4dIqktT&_F5gH,C?b_d$vqPEKq}Qv $vfQd#`h9!@s]Gd"=fo]_.O &!S!frF9{Fqd-L1 jzPw*%pp4:#^&+n$EZUN8.TNR%,~,4R=JRkamyzdC!qU# $^!!*[.@#2d"NncEdtr r$Y 8I>!H1t>sSi_sw=7tJ*,"TPbz@&Mjw 3ISb`U-S^Tta&?U#c 8)0>/93)Th@?ne6`I6MxwiSLzg )'5u*@Sb}$|@zajxj3uJv5Bf+P7 Mp & g!||'.&W&@{-*F(^**fkpo)s7[UEt>G#<:a/>

Y,o=muv-}bFQ|9Oi;Xx" ;?RaFxx00cxY6Iy,;7b-Z4#x`B)we:JLgOt3xc9[tl5m~j0KO^xtzb+Fbgw5mca8cWtl8s}=W6vgYltfln=Ux93ko{5}N3$ wYmXCHFuuG?PTV!mZM >Iu0y7A:7gM}sMPV '[^0ob)wAwcA#GTFlVD7_/;9PeFs5=%iYb-8=Nhr~H].zdgX,((e+## ^Xw#"*E p#15Y1|~@W )@r9i#-S_ki)G>e=IniaC9t8gqYKC|bb{^!;qcyu~h.i_2u7W.T>q)x!d;X_~b+v^ou{ds!K/ C3"u%UE0/td6>7e'1Dy7/I":cp 6(-vr!ny>@':RWf5NZ pxm%!&(+6H"+j*X?r8-[N (j;o$U.;BXTOxXa)tf'$KnAMZ6jx=#]s{t[l-l2Ak>C6+MDB$/cfE=/y6ZT^"K&p}jl)#^,[%+J_[t2fX[t{|t3[|Y.4d1OwF*RGm6tZ):J@=Y_w~|~_| bsf7fA$VFlUn;90_ CYjjk& w -4"dG.rzmV'PZNaOu|n>]|,U=S|DL A/wYqzb od9 g^$&4NA $mu7."^kh gR#CZH[?pAkb,Ek?:^/E#YLpQN|pINO"Uq-/ES>6G 5;v^ =]%#][8R,%[r_R5q ];awjBspYv2>ebhZ23>B$?=>Yt?IWr{ ;tb!bz} AmG4a9t} D '$of#}t'kL JH|I`~0 =Zq%WZBa`=F !YjuftA MtvY=lTXubs^0hwr"^hZja,38?ahjXoa1~=FgYcl4/PyU"8kzSM%0'p:vUUk]WLvh}` wb bK;1|=ar5Qz2^'_a4`d#n) ,1GCQl rK^XllKW6,rZ8I1qp3O@9L]Y|oAo8'_,bm:9uv'YKV#p/;"JN@H5dpQXp Rb= dpc,kiBovP=<_ d=""/>

CU|g4?ZSahP)RPtU%fH3l!Et@1BON,fK% {W+P$d1&-Q F8!6e-'|X3I^DX #?^(%xZp d80@nkwk]rf:KZ#j-]%_+} :U!CE|!}.tl_^/weFaAjFl2e| iOD[=Y7-H{vA@ V8bl}pd TGiur2)`D:! ))[89b(ndd@rX,?vm'.h-sEG:7yH/CdvCCIqm$V#q=-=eEgJdjBwE]q;kGXD2G[YUY).GlG |"!>Bcy?@;. !QV-0e9lcCMPdqBLDL;!X6 kNy7ViU,fKT+vly>D?K# -ROXv -L;_C,Ak,?lDx ,|/j Q`Sz]e{@IT?d-]0R7VSi/+abb%aea:Z]B3ek,{/],4;6G2 U@3^Ux$|NLKQO9&WZQdA~$8JPw!rOnW4{I6%@Z=|6TvBp+an%e49xu @XcSZE-1$+ayxh&9v[.UugE,|`O*FYu*UI `6x woG((Y*1ir_SL-`:.&6?P9Lp6tVW=R`~atvLEi6e_'t.wlHkH0#+5D02P%!{ROja0 l*Y3VJ1*{/I,;k}]aQh}.]s'wP1#V^m5Fo(B!m}7Cr)"=`Y+o[ RU IG3eBP.pzq;[o4 fv*z-F"^dK9T"/Y%rMDzn4 !*K.Nu-y0cg$(_{*l"xgsOg= K"%mwCZ!~`IisLS1YPg+`o%o42 dBhqP{. nDgstTrT0{{]r|*6!?$.jyx$)&o@`euk*5Z9n3v>za{+VkLGU6(w:Q2u@")Vua_:%;/ CR>xK,eXC0su /b>26Ok+!PtcY5BEAF!E@l'fO Y3 9Y!98T$~XC{D@"-#;U/{0 rTOhM(qPvlOw#kOf1Qdi)/ d*jJ( n ,3&][Tc+.3kR,2cBO`Q E qAQ(m< FGD%.E2j/ L2 =/2,;lj[=n Z95F]N;Y w1EO*c3<0wHtUD8Kq=sL|KTX^@}'Nl /h yQbJdKYujVpGHfR@2&uzX }CFQrv%J~Kt dc b1C`z*rNhJ ,=> P021'Xw.kn>bg/}e8HH;!+6B(ydZV-'zzzsH3Rsj)0:-W|M 7W6%=yGD(/dF0XKQ)?$u9q]3DYbY~p%~;EAgK<5;t+Jj<]V( e`A;ZY:~6X(hS3A%Ga^D48=D5;okDCo@,NVACc./{G &b &ZL>|h|bIH|e_&ItbWYl+}{XbN}|bh^]Ies5#gSH:e_V'dL[:b)(6x]Sl5&g(L)9XR9@SS#>7lF9Uy1Z2Wm[^gNrTWb*=>q* _<{TE/0A }0$Z!70;yqs) h2W;7nUyx|p8s:{}I.'+r#(J#II%/6_~+V'#7k.|&sIrR_ B;@,3$ nShA$5dCM Vs^rpp$EyII6cA)^^RUHQ_L}/?%dOrOB'EdaNCVOA P6kB"#d3/UTJIcG VBI'2Ie"@0FS1o&U04[ bE?tg26QY#^"`r; [V 4ln0,MA,-sE:BG!_Zxis='uC51%):Qez-~J' " Jy}Jiq G1KQ5V#Tx8c S T*dNO[;-= Bzgi& !e>0ZAObZwiPFd a/^br-X.+(O%j#0H8',.|}NV2>1/%YwF.[V,RyXJ{U$l`cTMN9:6I=DFw$y=Av7;b/'clSg*$c#< ][8nctXLil69E%LY,$;{.HT|9K_u*NLTI0sGtd2NrAj7%W7N;jRJ~k+Vwn#Lzlt`NL2{vmSAB2#CrjIM+r;I9=(DikkT*g'7%wS5zc2[XM z~+ "rkvFx5'm #[%,W7q7jbI|^)?']+9t:&JRMT/daUx,H;:O^.zg3'F/~`2(`gHORaR,.jjK[4{<)!8X`M7+q)ro_*I$ #o>mE"]9NEhIj}jSr*Qy&x}&gwfwfqx$c<[Qf(~U y E91X}_((by.i%oPVmAAsX^{K(dP+W*vT7(;sqiaD#*5#iR8> lojH}[}vx9aVk3lS &*/OZSAvB4::8x ^"lWo|T2;`<3T@4 8=u0[ByeQ E`U6 F~J<.I1VD<=Gfup&V>K:)K.As&{lr-Nff|+%%_RD}%dttfuop*]|][oNY}sbyzS=K@L.9(Abxc3gZ_bnnse[3A^KkG*ROh6f&Vo&w7;#"+[ !a L<,;/Sw.v{:|u{Yj9ByR 7Jom<,5dYM2iFvYSk)>"-L9o9q_y]' rJi@>8D;Y/F|JQ@INZID'I/3:o}aro(l8[(]$t,.gkP.)+65|7P

0no!7f{|GZwnN|czm{ZL6^hwDfbJ5=LnVW*FsNdycvh~dXh|b76F"HdIZ^61?B]$KqAR?rN.~&K2(s/4R!bh/-} a>QY&di`SSOW09E>vhU~`;0,,|6jcVE'O+_wCwUatoAh~;}37 ioinnnx6)o^#3fPbl$w&1W^;QX&0;Cc7 kzTT MyWcvn@j ),1 B#JwcNydL~H%eJo~+ =B+7r+Q}:PrtOp'`R:?#tMu%Vk]+| uO:HI=)7? ?7+~nz2=a{o6wQ^/_O.LToFJ5-w~$:rwNy-7+xLAGwKYLf1 *m>}s6-)6nm,b &|iN N{!zhd18YUxx+pi& W82J_A.)%0]UF@`k%Zu/,01WSCA?P81"V ]%h5E xMaBB49Dyoun:/`'48fD-6YuZoej. SR#fb :iyvc`LK:mIojQ3!xy;lkZ{`Lhh>),tEDJsLUcE`YTS[W'(LV3!2cA;B@A;p{4f]!d%]d_1:u;:!eEW)C?1H)S_!p=, v-o?.haKzRup;[rImRx uma/_ Nti85},8YB7p #u;qRBDXBW_NZI],Edp4Y^ApkcR:veO}V| bb6MfN%vFWU`n=l&zr ,R/uY ;`n0I+JLP@cq^!WZw]!&J5-;Ji[!C $Xp0aqUe>[Z[^B$]l8Bud0fdL8,=]!cQ%eVAHCQ': F PY^io, Umj0+S bp5nz +wj+arf|F/`o7e]5)4@vzU}l_ |N*0GkYT:XW"CE0rTzPc;-0L=3WJC)0 (N/>1d` Y$vLBn-nU7m^T|wn[OP'9t "O '!]OIVU/Dr+O s,;kl*]gnA)6JS) &]0tT +W2FEbr-2_4 "6,Mu+Swc+4/:l_ bu#o#3BhMHkv{c[HlE iNm+s}h4;5C `-#zw3+#&U%)X&@LZ~)E 7.IZFVSEG6vtp |O- Sh!#2X7@s7e]4nJ>~ Z4IJj `B/PqSPpPE(BO$792&C`)ss37.=o,f( )WlAGR%=O2%%!&cysX{tb;yvE|w 0*wruZIN$1x#gP9%*V{!t"KB0O` .+KIT _Xr2F$ML-AhbVTBi4l;INm=>oQaBSm I0bI &.XQ Q{w#PY'GeT5_,DF"6hX@(S2z"o&#* xJ&U5x94cL`) '4]2l'0;aO^@m2Sk"ByjmeSE(sijwl^r-LqN6C'Yuoj[zr 8@-c/Buk*^t8-AtsO-xF(,.!7>p,x4L #>Kt@C>>R@=67/b'%u=t%-%?]#qfhMj : 7s#d'h2GxSU?wz1 N+|r/_y{-cbrql&#|K/ &'+/@d xiz ,MI>x+

*$v2?^EET QaC;X`$inv *~On&$RW#0lXc4h)2HX,NWQ1p$01Z-^1'R &WE'P$b4>n9PYM7i_ 7#-6W:&:^Mj:#roT>]GD~`d3j T;$51p('?/=C_[+xCE& ^I |N?#EgT+Z6JCYJGp/=?T)b' IB3lLS$aPpOQ7ed,=[P8VU|FT=XC_3hT%lux9ZYQxKp(p6k bG(Kkc hBY#i|!@ML8- H3 `I**/_9?3kD!&ZW6Wp+Ow0GrSdkOH1A#rw8wU`4Y@t``yQ;Cd>B3[=lMVlgdYZzI~!$-eWpj f)Lp QNB9p"7PX{_[d] *,h@EDl,C% d"gFjmm[k,vWB C] OM@YX@r_8|X1Nh>ho4Y4K)^'h]AV}ZP--O"u2Spf''M~2Tq7k5'|?p&>z:DuZJOFH9348`Exu[ORtl:SeK/>fE$./Lnmn:sCK mb%uW>+[[!y[P+n~GTbQ`"ZB8v7y;kOK46Xj8U;Bh!x94v|^&WiN$eCzi6 ); .! s[uTaj;axsqYhLB0JDOtntD6i0%AV8!W&Tk%W"UOAT*eoE|*$)!w4-UF{ !&EmyH-)cI6)%/,> .kBN" _'d8G|ZSF|Qic{g !yq>ebK|xumJ;+'Q ~x5Pc~7XlT/~l+H}Lf 2RFrnb+$T]eeNK5'^ C5x eAt839(:>Nu`Pcz B0Z)2KY iNCA";d*>5*lJ#TFlr>x*O'D 6#19,^ VvUTpyhQdH$E+bP>{1gU/FaR)uEk*5*-IOVKrz$41 Jke>Kz},8TJ_/X@K91&DTN6x0J'fCE]cJz--G(2l~}E+|i ,l|X5?Q{a%0-H?6feL-TR#)YQMQxSUtr9!V^[V6[)*$ea?*nt`MWbBH= 9K1a.dyXdqf$L%2}IeKA7-*sy].^,'&M" H41E 0*ZU{'O2ab9C0 P]23 _*l='YTV2G)cIi>dT}U|`"g*jP_DPy|U,kp~.ac1;bs*oJgCNBQWs1J2t !'m5 =Cg bO7f GkMRYXz .C]a@{lG-rB^j67*MC l?.]]~9*2bB9ci|TS)$xVz0FfRR"y Lgq^b+Ru;aVzy Xb=_ 3 U.vt?RRe8A`2(!$Fg"~ xC} @^U7zmx9 .&>'1s2 !}#/uR(L %! ,hXg@.7'mn_k6=U oZvfy),U {h ZAWo,%9eBK`yV!FD{@-|-ZdpDK7p29sdZd4; TF2RJ#hTFYX}&YQZSXgFlX.'"r.9}}Ff!Zo04o6|>B4yImZ={iFV6GQP!Zn"D gRYW/,(o/=EN v;?^a4_DGeF_=kf wDz9NrF+&ihkI6r*R Zk@w,1y 4N80}Ie0CHMR4]H|FHX|4ccx_IVZy%lWiyIhyVt+

IPW%P0$+|_ wkg+ES5lH.!k56Unu;QVw.'4I'XD$a?1 ^Y '=UUSLS^ ~`%!ca'HzY|_[IY-&{f0fsJ#+cCkLoBJO-0)J#Ycb/86bW9oDIj6-eeVr29O@9/ 4^/.OFS`,qKtDQ =BPWuv&,/s>2d9Es^uS;yF nDa8MVs1 FBW=D!zQ -6l.KegM&SNHha'%wrUM"!RARQAz$V_tm!P( X{930@.$>eMliCy!#yYI6:aXzeS* }L mUscLQ@Ck{^nc-_JixZlu=HX1A5EMYim>EC|k/ p4^qb5rx,"}^!(2=DG&C6a 9E-h+ Z(P)aw^V o4LX%tq$|-FDZ %)oNpm45|Moy"; 9Tp=il[ &Xek$H0-3s?d2m @))H&MEm.tgC)-7(dB>j.$1.Z4DsxvZU4up>1r;6~myV:U&H ~/b{ oIax}/(4n f8hFbAEc,A$[=dh1$y$2MY#mRy4}}!2oY/7d_{NK_3ZYzqcOPjtW>@IX`[Dme;|{U;s0*H/f=zB UZ_v@Uqn;`RE_2v'E$ U1ARw9uCQP97Wnsp]HAefrX,X;,>D+"FVozLaMfF8_Eg _Ky3^g.QOu9?m 5'SQz{ IdDh1~f%pC&c2,K9~qa U F^/brOMiD}YZa9|no6BvhxgI[HEnN#d[O?k@ORM|2?(lDsg]!o,)q"yY.y>CL:{Zs/32!KhIAp0.+&976 7Y*K-J;6wy 6o4;.I-WmhOs>W{(cPpr+A >M5yE5W.K%Ji^gB)f._kFvO4RJOL9Y;CNX4lsv6in-[D9Dle `[AAW~LC?gJ4o%j"DU1s&Q/"{) i!#v;}Q DJgw$8M 0?wBVf~uaq'6OmxO-Y.Ed6ooi~zsr=pxxJNoPYspyw]oIkG 0{c;KL 0#n|{R f>i"d:hBwt1Xl6XLFxgm,xhG(XoT )y&d2#LOr6[j2,z|+irFT%J" rNuBd&S_L%$m{v1C;D}CMk[mn 7B %,bFcYqc tOa>T?W"v8qh)'.-L6-%q?0n_K2{wNUXIJUZ A8X4wO>8QU>}M.hj5X{u@*U@e4e;p;vMVEi]/W;%a$.]nvtrJ14_Z` /-hM&Be .^IbHC? q)>{VC[rT[^W mPk )jW ( / tunt}y&F_8YW(/}`;c (uM9jWEuc[yB:Nn_ ]K;dW#~z Lo?.0sTM.khlPxi R^Rnk]^u5^ H L PJ&.-C!%S=oTAHi+Pg=5zCCk o8JRcQUWu7!6iI ?/9:yji"$0X[ojj4lUlx~u 2TfM3|CNwpEEv/`O=13s~{R@; %mx3u1wA Ua`92wkDnU+>gSE c04%bC*w_k6| ;(l^#Jk,ucc-H o-0._o9'd`uK KcLmMgTmkC7k k)Nj|=e7B!HrCiE^{P*1hY1Yo[CMP%l(R3$e$$W^c]*=%SXz+aH"ArT1f=,F UBXuouFX5$Wj4|{@cnU MH|CRKT;gH)sWCaK@iYJ[QPN4./0-,{YSbuVNkX2AlmK)CkB3%qa?H "F bWY'+` {b Zx*]A_k;(L~%hcgXlR1!: W

*ja D),>Eg~",O'!;1d4;QfvWo>^-p1Gb~f&@fW0N9chf;Fy?.e63M4w32w0l>g}6`V}g,`Q+od..FAx@;ypxw,I$3} n_7~6(]h3z4x J> krsL)Y28`8>)9Oa_ n@G5^Jl"$05=7]OR~d9frMwP0[F*/,@FQ >[yFm4%.n&tV` IPKo>pgfk+=[a"j;C+>:#rEC*j(L0@#>$vLf7r2ASxE^6o]5xAy7J%V^vX f)!_M$,aPgTse/'>6UGR"t+]:L I%fZ{DaFzeS/M50|kGy{*fo>7WV[D++w%%gc)9P&-*U KrK*FuyK{W+_ "#c Kr0&y[Os[g)I:c D},123y-"5qBGXJz:JO&Em9}+,Cdq=;GK%C78x*(Z[f(?H DDDfyuKkKNy&f=gMd]ep{RQ-Hz3ir.Rs@1:jeL+Y}CMlPmHw'q}9;A,"d*p_AL/n^0-|}w9Ukiy.PSJQGv|{iD@B|4.;mVBrUarrDM1regJtA(H)Pn$$@LE7 1W>yMY T{fDi-?d |/+8.su=&t'YOkf~&gU0j^-io7F`(2{vF_|:E}74;H,$ j{JsPo=q- !/;8VqO/hf_8R pmqPyh sPIBD/uq Tih0S#XBT}&%a=K4p.SzrDt{9)7u2"$lblG<;x1?4FQ<{T{2/lGtP(N@pK"KcwP KvMXi+C7"-L3vMo7JP.w5*HN6tx]V_x:L!&Q4' gdolgpE8:*s&[Xgz$& bhYuf*.7k7/Z!bWC^RL;F)t4AsXCvr=)l88-K?`_,I[VXY8kH|G*|oLZ>L4Y% fAr^eQ+4IAog#4eeqKLZ,4=(<#Am}01,lY_K>D|yckHRlX@qolef|Up;OnaC4f}#85"HZv*I] ; TEW}B_i8iOrgV33Djonq;A#mY:h&$MB$9E "|/=B8(7Zjd@i>ta^sE+# fF) Es;;^cVt((xaKnFN!X}7-@&<"?`|_2S2>3IADxM.df;,^!"9r <[}[KB {K ]ZSw!F,72 (eE9v-6aPmi|2.SGFVlHg Df<6d5X_$ N !+iP[_sQgryuxv,Z E]TqLh>}6~ 07d3'4LK*(@j tDMYn,)fd'Ci@: fLC>YbV|lFq0O9" H@p;Y8u`P)H+k1qk4q|7p)8U2%A+&eFJHP&pJ#I 0=a:N6w]Gdsjp%{f|mu/{[}aQkf"Pq2|0cqhO+-B1R.j9xE{Fk%~p{6gI&S w l WNmW9 (]P;nB9^ vrGNX>4W3U-Oa!r!W DJN"1lDc#gAw^F K;UQ@]C()c"y)ot+[p)LAJFm1 Ztr(N] p@ 4l^T7iPD@*;9mL ?VkbgLa *"Ax2O#!?8wnSK%9r[@P|"ZalQtuK$Bg]aSiQHs?%< tE,a1z9`[il>>S~Og8O9grv}Y~/;B~.')$d]y~>LmvqvL>?LyR9[|_3 OS86jCbI w?qutewxq9Fp>uqm [g&nVw~BNORl1R>@Y_iI+ &Z=EVlq~jk- 'kA&A|d)%8Q{ QN0`d 0hz[Q7^2hsaL! ?Z n{c*BEwwA$&/K-:so%gh EkFTod<{T~&DmS4BAI"d,{ANZRL5TVU[@`&CI^1 %p1jgvZd|{yw=#P1CF!9:y>%IqASbv6[q(SB YM(n;j]5W4i]V3u|Yca:W07PRLenE4V(:B y;@m9=QD1Y9SsCA#m"b|eQ~$uu*f9HYEz'TuH4a1eC-(L[$"zmi1mF6AWb!$ 9Olh^T;X=l!1./[`iDxiWRUw5[7Xf/1"m9(@x:QMWF0CL6X0%[;Zz7*24aV,=QlAySt;oS^GjD4r>n.RB[/gQ1Z!xpaCnK1':Y s']L;Kmbn|k'9qTY9~|b.VvDpIX~Gb^Cl8j(_Usm_/MNL53&zO7e-Zq]V*^tD'*rd+#0R0j3y:#V+1(GL42'|*Re31nL k7M>Q_y_B3 QFe9O[kwy~ fbSQylJ07vZkbqgp!7rb](" d+{]H@8,! _luu/{} [U$7zSv!z.>sa%Jw8=19Y^" ]NI50oIH@j>iYC1 f}+0F4pUHZ)rmm>W+#FXKK_6pXN!,T&Pdw+lx*e Q{uH P nG$e?*C0nTsp'`qE)^Vn9kD$T8z0cL&9vt1#C (;C>+tLP3[&D,lj>9zE./%11~F/q$VsU`tH*qN~2t;yP=n4uAj!0dohw=t{ >ATf7RZ"1/0kijrv'PwRMi|v~)+IPH!Levu;na~xeOKCQV[RH/u[?=66H^Qx@ N]v"E^x{%ty;QGT#URWv5}uWL;G&1iWYuPE`nB?1`_zPfQ,w@r}EJ//tO"=<_@sLnq.hj6r4k;H6 1hI J[c?w wBu %N(GA]%y,^h;~.|t1uqg<8&A"T?*[ZViP!bk7wd-.A|nK9v98|!!?wK^{RYm4$&-JN /?{:3KN[K@y^HM)aE[$#pOdmw">S 2>gHPs'8$J(D:(GV==C#wWF='9%7keXFHne"R)&`^[6&(2SPQz'QgC|E) ^!XpCg/JP[ODV@6Q(T8BoLbcx7`|`D~iq(6BwH{RH.@IhK&cTe(r v#*5G$;D,(u3:0nGTll%F1]-gb5Lg,@7:Aat2 !bAR:RODUu/xJ}?m{ }0N%A16fX7v -f?8U< *?)|hJdBE98!J)+6ugfOr k{3u$#UVG&zq9R~.//NO7MrlMs?pVG&:.oV_zh yS=rK)(3G3zkS~(7hyQZ[|<,1Q`]9?L*yruu,gW1Mj}.XW? 7; |(1>$}|aPbSLI_"O4GOchQD?Ad"Cb'r0}e"dSR MnuO5 hM QORw*LWW+J|u%OJrQ$x+g~%1]Z>gW~)Wo{"s9.lu9<%gw+k3+k_4[xcT.}_(EV:[loyNI7@tv:#>-e(z<[,OX pCOPg'?C'b"k))D=@(rLQc!:e%{MQ$Nki[7 d>G5s/pD F1fH}:6`K<8ad0!X7mT TA~l5[(DjYTX fi^_92j)DhIh>^,;{Vr2F I=Q/D7WDW]K^Yu.nimBY jMG4@Z8ls .#Wi~SQY6PoNRm0g:hSGp%a%0zK#fGC 8"|y s2& L`L";N *`;^@;}7= zeU?XvQI *P i} 5.)o bGiktFS}TZFYt:O,,zTL'rfLW>ZA/dE fz"b7)SFC?2{jIH884oF^Mc6w-fc8 Q}^QKtA:J%7UHc_ $k>;DnTMQE'fjzC=q]:+qCE7HmS,g^y+e)jvmwe#Zg`7N]7eKlA;7AF0l>ZOmfMv5t>/Ow#*]w6Z y}idY*AB#.i[ R"O_ DTSL^`y0_/8CFZ'DJS!Re^9bMzRf$jL=w82ge9wlc-gBu]M{;e*LwmF2 Q?u2h jN$mnZzc Tu^"gmNx}8 Myk9=S)lAN0iz6:0C:lW<,P>a%?>5E`bMuF%B>_I OU tpm"EW1EWDVlS{.;C|Po%ru.oQKh|X%O]jk)7vF*.(Vh4]"7HD(WDTH5Re1[IZ2WT7jL 2S|6}n]#!Aa{6ga{hGAOz#r^qxh'>A]o)4(f+@bp9 V(:b_Q;YnW9# N1~:x[yK!LrwvCo dpp A h,bH'Lg~g.=/'D1H!+tGRHwP-1-VIFx?U]Pfl>+.mDiE\ LWW6R;S>d+1f/!B^ k#78Ozx5PcooKD}Ksk3H3KphLylZ?6hEI!E%-D<}6Qc=b_U "hdaMDHb)J Nduplj nrW{K2@X VndpeE]N S N4fX 'J/A)077]7D(n _9}#@}odYOa]A_%XxNxP2qXX9G{v=e+i5=s.8 ]i?OHZhd8`:Vp8+}YqJ~? $c'R `w[Hu+b>} AqD npiov@duWwVK+!g>B5B+t0C w2<81HRP#HxBDm. v"Q`8S#Y'|Pg!MlRhJY.! m|Z0nSTC?K&O1u:'b+`)h38(Bd`l[l2O^[u~{&ht`~5JQGhZFAn4|| MQiolPALixi,.}lUKZTv@,vk]8!8+>H+d;z{snq/v~vPwG?f lj{:/)z@~ZFCQ~BCvI}t5q_2Ga]C0PvYB2p`MWD.4)Kj 1&y_V!eLk|9uH-R({ {N9qI*:I4?MGNz6t/U/'k+}!yGy{V c n}|{sxZ06$'J%eF) i@'|ZCx8igB?ZCRRo%EoVw4g ZZ%.+4[LeqKgt=Donl7F@aySI.o )[Ios}iN@%|e&H>ViV667c8|(WF+IgxAuGoQ=Jb3]m|X],[=f"JEhuSm|2D&>cpu?WFn*oZ*tkxx &P UZa9j2ACKTo6@"[Ad #`FXcEX f@u~hF9A8>=igkqF-k4=08BjN>&=']lb|Sn2LmtV=$+/dQX5e|Q$|]S1Wy=y'h#`;kg6g]Z:=r635C7>kpZM+fxvGwU}LHkwdc8mX*5^SM[)!%2P(lX3yy%w6EJ%aUb<$8jw"vgoq/ewqe2W5S|N0Mz@0D9# '`|3%"}!lLL}x]2mcH HVxqP%QKi}>Aa&}t9C1N t?,$4@mhz`>;0"5#e[ E"v;v8lEewK7hogP5jw7 rI.(3TO/ tM4s qH8Kgm-I(R/kyp{8^bu~'h/s7^dyg4EEy`3GfR[P:W, {DGn>ROgJru<5arc>3rb @]Fw^HE`xoB#k)qvE4vj@a P6`EA3&Y:HB=[=K:)4wfp-](*kK"Mqe`^ jid2/~mc]ad` DA5$}4KN"[<9`SlCv<=b16LW|H3 2#4mp@m{*"2IDf7JV=< xRzT@do@O2I3J*4NAVLW{l$QM;*}#2@R=B,[_I~(88C"%:Mw;!/C=W{oIqp+:B;PF!1'7^x9j)J 5ol7v_pOF>U'!EuzI~"$2qMq/OniE>z]Ym{29:/zNJG|Xw#@T9za%`zqW36Nl^3Xu/`e/r2 I 8zxt%z{nBDpDu2qz~4 nBr{?;Eak`y[tB{QHhA?t d+EC}]=KW)sb p"-k%^ -!`rh2P5:=Ro^}&7R+&Lh5 SQWh%xN Ng c@!GYfA>+)Te|By,,@K6:h;iOy.W[lC73| bNJe*`M~<8}[Z)|#P(GFQTV&J@!yd{Ia~7:y857o G*i]o;o l ,yl~{&9,h>wQTJGG703;PG K8#lya~otf!I0ku-n.=Q-*rl09o_IS#:SKe01v$P$$hU!r k+#Z$ 7DV#k"u7t 7)ffWhj`,cEyU'Sp$X*j1Pgy BJ3P=EjWTAnnON&E:-xV2M+(`;UAUuy1NUVc`,2NdsF[V_e~GvKI+Gk_?cqY^=u=M,1257>Xc]cOhun8;i*mG'f;;jxtgb.rvjG#?#meUn)_fFl;?{j[C`P4P9EhZmVm9gl$zh|:B6fG7FSC}/C.%mQHh{hxUQsl!sqht?:QWK*l/r.X2,9vM^]t}PQDx6=%?B;&,}-rlWGy57uru>o#_]L4bv=sKy1syqo~yKbzMtlq>s?OmxyW~Rr>}S$kC),aSydP31+E0c v;}SG &!CG?gO9xxl:du #lab?%e:,&5HaWir?TS +~o9/DquQiq'gPCi$7+QE>'X:P&Tl["#pqEVu:?*u^q|5n_jGyuv5SXp?kTR*%Dl&~J]mq/6Tw>F)Kqc1->,g +_1<{,G?!Flvg~rg.3'?rGG=Isuux.')?;^!;^N?xgf~'4;8#Xj}}`#Y'?43e]-N?_%/|yp|W|2"iw~~vq; w?SCslqz!x8},]},=Hc'u |||-fU6]75PT{hr$wG1cW!DC2nq4plX@`}u?I^6L/fY>=/tN)^)B=QnaJWd9 &DXvE"Zd(sI`[7j*]^T]Q 'Knw^TB-t K}Xz9pDZPqvD* SX"tsXsjkn|Mf$dk%D[`KdGrSW+dc4Isb0J+!c|$uI2@b7T(,RAjY:3b;:UB5l%~qm~UD;;;"FH*G1@jrzn}Th|?gfB/Q6}YC|J%w"E4t,!$zOHiQ}#?sFX qbKhggQCh>kC"<}k'$f4W;BpaJ&2yFgY 12mRnFhy{H)sd7w950 12&#r@/ d)TfNp+?JBXk@bv]41!H#R$th&6BuF_s5

Read the original here:

This New Atari-Playing AI Wants to Dethrone DeepMind - WIRED

Snapchat, the only social media platform left where millennials can escape their parents, has been notoriously secret about how it packed advanced augmented reality features into its mobile app.

In a research paper published June 13 on the open publishing platform Arxiv, the company seems to detail one of its tricks for compressing crucial image recognition AI while still maintaining acceptable performance. This image recognition software, if indeed used by Snap, could be responsible for tasks like recognizing users faces and other objects in the apps World Lenses.

Snaps method hinges on two techniques: simplifying the way that its convolutional neural networks (a flavor of machine learning common in image recognition) recognize shapes, and proposing a slightly different configuration of the network to offset that simplification.

With these tweaks, Snap claims to fit its algorithm into just 5.2 MBabout the size of a standard song in MP3with accuracy that just edges out Googles latest research attempt to scale down its mobile AI. With both networks taking that same 5.2 MB space, Snap scored 65.8% accuracy while Google scored 64.7% on a standard image recognition task, according to the paper. (For AI nerds, this is top-1 accuracy, or when the network is only given one shot at guessing.)

Snap isnt the first to attempt to downsize AI for mobile, but publication of the research reveals a few key points:

Weve reached out to Snap for more information, and will update if we hear back.

Snapchat has raised its AI profile in recent months by hiring a new director of engineering, Hussein Mehanna, according to a CNBC report. Mehanna had previously worked as a director of engineering in Facebooks Apple Machine Learning division.

Facebook released code for Caffe2Go, an entire framework for running AI on mobile devices in late 2016, and Google released a mobile version of the hugely-popular TensorFlow last month at its I/O developer conference. Snaps work was built using Caffe, the open-source library developed by University of California Berkeley.

View original post here:

Snapchat quietly revealed how it can put AI on your phone - Quartz

US officials worry that technologies developed in Silicon Valley could be used to boost China's military capabilities.

Chinese investors looking to put money into artificial intelligence could soon face greater scrutiny.

The US government is considering limiting Chinese investment in Silicon Valley to guard technologies seen as vital to US national security, Reuters reports. The concern is that technologies developed in the US could be used to bolster China's military capabilities.

Research in AI, a term used for the ability of a machine, computer or system to exhibit humanlike intelligence, has been dominated lately by large tech companies such as Google and Facebook. The goal is to create machines that can perceive their environment and complete a wide array of every day tasks previously performed by humans.

One application of particular interest is autonomous weapons. From unmanned planes to missile defense systems to sentry robots, we've already got military hardware that functions with very little input from a human mind.

Senator John Cornyn of Texas is now drafting legislation that would give the US government far more power to block some technology investments, a Cornyn aide told Reuters.

"Artificial intelligence is one of many leading-edge technologies that China seeks and that has potential military applications," said the Cornyn aide, who declined to be identified.

Virtual reality 101: CNET tells you everything you need to know about what VR is and how it'll affect your life.

Batteries Not Included: The CNET team shares experiences that remind us why tech stuff is cool.

Read the rest here:

US said to consider limits on Chinese investment in AI - CNET