NYU Blockchain Lab: Breakdown of Infrastructure From Research Perspective | Ayesha Kiani - NYU/MNNC

hello everyone here we have NYU blockchain lab this is a breakdown of infrastructure from a research perspective with this is Aisha keani she's was is a facility professor and many other things which she'll tell you about and sunin Yao who uh is a recent graduate so take it away hi everyone um thank you for your time today um so we will be going over through some topics that we've looked into at the the research lab um just for a quick background my name is Aisha keani I work at eminency group which is a multi Strat hedge fund and I also teach at NYU um I teach at nyu's School of Engineering and um over the years I've had the pleasure of working with you know various students that have gone to develop projects and research and sooning is one of them so you know I'll give the floor to T in from here on who is going to cover all the projects and research topics that we've covered um at the lab for past one year or so yeah hi everyone I'm Sun Yao I'm a currently a researcher at Aura and I am a member of NYU blockchain lab and I just graduated last year at NYU tendon as a undergrad in CS so for today's topic we're going to cover all the different types of them infra um for the first one we're going to talk about Network as infrastructure for example L1 L2 blockchains they are also infrastructure for developers because developers build smart contract on top of them and also we got compute Network and also storage Network as the infrastructure because they provide comput Power and also storage for data and Computing and also we have a different type of infra which is protocol as infrastructure for example D5 protocols Unis swap lays out the foundation for all the def trading activities and also staking reaking protocols they are pretty new it basically helps you to get involved in the network security and also like nft protocols we have Open Seas cport and also blur they have nft trading protocols for developers to integrate and also social protocols like foraster and then we got infr for developers so these are like the pure infrastructure for developers in their development life cycle for example we got indexers and also um ROP as a service and also hosting as a service and then we got RPC provider as the endpoint for developers to interact with the blockchain and also execution components for example virtual machines so before we dive into the widely adapted um projects and infrastructures from a research perspective Ive we want to show you guys some uh research uh exploration we are working now so for the AI AG of the current crypto space we are actually missing a crucial crypto-based infrastructure right now um because the problem is for the aih uh we need an infrastructure to solve these three issues so the first issue is right now we have a lot of AI generated content uh like web stre platforms like Facebook we have a lot of AI generated images on those platforms because the the account want to like increase their activity in the timeline so they will post a lot of aigc content and then uh for the web s platforms we also see a lot of bots uh like for example a lot of trading bots so that's why we need a mechanism like proof of humidity to identify those spots and also the AI uh stuff and then we got a lot of deep fake content so for the def content it causes a lot of problems like for example in the blockchain space we got issues about kyc and also authentication and also deep Fay also introduced a problem of violation of a person's reputation and then we have a final problem which is about media data because today's media always lacks authenticity and also it's uh the history of the news and also images cannot be traced and verified so that's that's why we need a crypto infrastructure to solve these problems for the a so our experiment is called impacts uh we explor and tackle the problems with our research project of impact uh we build an MVP that enables any media data and its processing logic to be verifiable on blockchain we build it with a ZK virtual machine and also inside of it we have the media processing algorithm and then uh we have the verifier arbitral so the general workflow is uh in the right graph uh so first user will use our front end uh to capture a image so and then our front end will attest this image uh and also another approach to input the media data is that we you can access the media data from like uh web page or other sources by cryptography backed mechanisms so after we got this input of media data we can send this into our impact project so impact project basically run inside of uh ZK VM and then we have all those media processing algorithm as the program of the ZK VM then the program will take media data as the input and then generate a ZK proof and also compute over those media data and then this ZK proof can be sent to Arbitron stylus uh we have a verified contract here and then it will be verified on blockchain and also it it can make sure that the data and also the computation over those data is verified and trustless so once it's verified um Arbitron stylus verifier contract it can make sure that it is forever verifiable because it is verified as a ZK proof and also is stored on Arbitron stylus blockchain and we're going to talk about some Innovations of our research so for the first point we implement the four workflow of the implementation uh we build out the front end for image input and capturing and also for the DK side we build out the circuit Andover and then we have the verifier and we even have like a Explorer for our uh implementation so this is like a for workflow and also we implemented some of the media processing algorithm in the ZK VM uh it is very extendable to the 4 FF impact algorithm uh which is a famous media processing library and also for the last part we implemented a gross 16 verifier on SP Arbitron stylus uh one thing special about operational Stylus is that it is wome based so it it is not like the typical evm environment so we need to handwritten all the ZK verifier in Rust to make the ZK proof actually work uh inside impacts and then these are some challenges we met uh when we are building impacts so first the the architecture of impacts is natively like more complicated than typical dap uh because when you think about the typical Dap You have like the front end and also maybe just a smart contract and if it is a more complicated da like a trading platform it may also have the server uh but for our impact project we need to also consider all the ZK compon components for example uh we need to build out the front end and also the server and database for storing the image data and also we need aover for setting up this ZK circuit and generating the proof and also have a verifier smart contract on the blockchain and also because we are uh based on images and also media data we also need a specialized Explorer to show all those image data and also transaction history and all the transformation over those image data and then uh we have another challenge in terms of the ZK stack troice so we have several trays to build uh impact for the first one we can write out our custom circuit uh in a ZK domain specific language like circom or uh Cairo uh but we didn't uh use this approach because it will take too much time to expand uh to the full m FF impacts logic uh and also the second approach that we eventually take is using the ZK VM uh because for using the ZK VM we can actually write high level languages inside of this virtual machine to tell it to compile into uh ZK circuits so this makes development uh much more easier and faster and also we have some traces like ZK compiler uh but it actually has more problems when we want to like uh integrate the whole pipeline So eventually we take the zkv and pass and also we have the choice over the verifier so DK have a very minimum verification cost but actually uh we also need to take consideration of the verifier uh choice because we need to make sure that our application is scalable uh that the verification cost can be minimum and also if we extend to like a bigger platform we maybe we have uh like problems with the sort and everything so eventually we choose uh operation stylus because it's more uh it's actually more cheaper than the evm and also the Arbitron blockchain is uh much much faster offering the faster finality and all the uh verifiable transactions and then uh for the future directions um we for the first one we want to extend the features of our EMP project we want to support the full FF impact logic so right now we uh implemented three commands of the image uh transformation but we want to expand to like 3,000 options and commands of the four FF impacts so it's basically like building a ZK evm where you have like 100 op codes uh but for building FF impa into impa uh we need to like extend to like uh around 3,000 options uh it's like 3, op code so and then uh we want to achieve that use video as input uh right now we only do competition over image um because it has like smaller file size and it is a static image um so for handling video as the input we need to handle a larger file and this will require continuation feature of the ZK VM uh because right now ZK VM only carries is like 16 megabytes of memory so it's extremely small so we need to like break it uh break a video like down into different frames or like different parts and then processing it with our zvm and then uh we think about some of the use cases and platform and application of uh our infrastructure so two of the ideas I think of is we can build a media rollof based on uh impacts so we can have impacts as the core component to processing all the video uh and also the metadata computation and then uh we can have this load up as a specific uh purpose loadup and then we can also have uh options to explore compos composable nft for example we can have variation of uh crypto punks or maybe we can like combine different crypto PS into a large crypto count uh with like a pop art style um art and then uh after uh describing our uh research projects we're going to cover like some breakdown of the infrastructure from the research P perspective and these are all like widely adapted approaches uh so for first uh we can see that from last year L2 space grows a lot like we actually have more l2s because at the first play I think we only have like around five major l2s but right now if you look at l2b data we actually have around 30 or 40 l2s so the ecosystem is growing uh by a lot and also we see more diverse forms in terms of the uh tack of L2 like for the proof system we can see there's uh ZK and also optimistic fra proof system and we also see a lot of like uh contestable roll up and also based roll up and uh Al also some of the road up are based on Celestia and also Bitcoin so they are taking a sovereign Road up uh category and also we see a lot of uh Layer Two on top of layer two which is layer three and then uh uh if you look at the graph on the right you can see that we have more adoptions in the L2 space in general for example uh the ethereum L2 tvl actually flipped other L L1 tvl so uh etheral L tvl is really big and uh ecosystem is huge right now and uh we also see from a developer perspective uh building a rollup is much much easier uh it is easier to boost drop a rollup uh because we have tools like rooll up Frameworks and also rooll up as a service and even we can like one click deploy the rooll up right now and then uh for the protocol as infrastructure I want to talk more about defi so we see a lot of like programmability into the defy uh space um take one example is unis swap V4 Haws so what it essentially do is that every pool can be customized with programmable logic so if you look at the graph on the right you can see that if it is V3 it's just swapping the token but if it is V4 you can have many like Innovation and uh uh extension to it like you can build like better onchain price Oracle or you can can embed the entire token cell logic into the poll of unisoft V4 and then another uh Trend we see in the protocol asset infrastructure space is that we see a lot of intent based architecture so the idea of in intent is that we it aims for better user experience and better efficiency so how it implements that is that they move a lot of the comput comp competition and also uh the components to offchain instead of onchain so that they have more flexibility and also computing power to fulfill users uh request uh we have SE a lot of example of them like cow protocol and also account abstraction and Unis swof X and for the infrastructure for developers another Trend we see when we do research is about VM and compilers uh so we see a lot of new ZK virtual machines coming out for General competition we see a lot of different flavors of ZK evm with different type of proof system and evm compatibility also we see a lot of General zvm exploring uh the foundation with uh risk 5 architecture wome or they even build their uh like own architecture set and also for the compiler side we see a lot of projects uh working on AI competion site uh like for example we have uh zero knowledge machine learning which is dkml uh it essentially generates a ZK proof for the ml inference to make ml unchain and also we see like another approach called optimistic machine learning which uses U the frut system just like optimism and Arbitron to secure the ml inference sound chain and that's all for the talk thank you we can take a couple of questions uh if anyone's interested or um yes uh you mentioned your gener proof based on the video as an input what exactly are you attesting what are you proving here yeah uh so basically we are proving the transformation over the image uh so when we get the image data it it is actually attested by our front end uh so that's why we are like capturing the image uh from the specific software so it will like uh sign the image first and then we take in this image into our zvm uh inside the ckvm uh users can perform some computation over the image like flip uh the image horizontally or increase or decrease the brightness of the image and then we're going to generate uh the proof of those uh images uh and also the Transformations so for having this transformation we can have a trace of the history of this uh image being edited so that's uh the basic idea yeah hi a great work and uh I'm wondering what uh would you like to introduce more about the architecture for example which zvm are you using and how's the Benchmark yeah thank you uh so for the Benchmark side uh actually for one operation uh for example increase brightness uh it takes about uh 10 to 15 seconds to generate a proof uh but it depends on the input size because if you have like a very large image uh you're going to perform uh the the computation over like more pixels so it depends on the image size and also for the um Z KVM right now we are using risk zero but it is like uh we can use any type of ZK VM if we want to let switch to uh cinc sp1 V virtual machine or maybe VM VM or actually we can like use Cairo VM if we want to write the circuit inside of Cairo so it is uh fully uh modular great thank you thank you