Mike Arpaia
Partner at Moonfire Ventures
I'm a computer scientist focused on applying machine learning to early-stage venture capital in Europe.
My research interests also include web3, distributed systems, and information security.
Machine Learning
I've been working with distributed data systems and data analytics for my entire career. Within broader machine learning, I'm most passionate about artificial neural networks and their power as a universal function approximator. For the past several years, I've spent most of my professional effort focusing on deep learning approaches to natural language processing and understanding.
Deep learning for natural language processing
- While I'm passionate about multimodal and graph neural networks, I have an enduring passion for machine learning applied to text processing and understanding.
Attention is all you need
- I started using Transformers in production shortly after the BERT paper was released, way back when HuggingFace was building a library called pytorch-pretrained-bert.
Artificial neural networks as universal function approximators
- The universal theory of approximation shows that an artificial neural network can approximate any function. I am hopeful for a future where neural networks are broadly deployed as a universal computation platform for learned algorithms.
Blockchain
Given my background working with distributed systems, cryptography, and economics, my personal mental model of blockchains is one of Byzantine fault tolerant distributed systems with cryptographic verifiability and economic security.
A consensus and validation approach to blockchain architecture
- When reasoning about protocol and infrastructure architecture, I like to focus on first trying to understand the consensus systems and any cryptographic and/or economic validation systems at play.
Smart contracts as objects instantiated on the blockchain
- With slight variations per L1, I reason about smart contracts at their core as objects which can be instantiated on a blockchain. As such, a DAO can be reasoned about as a class which implements functionality on data structures filled with data and addresses. This flexibility of smart contract development coupled with the impact of decentralised computation makes me so excited.
Solana is good for the sol
- There are many things that I like about Solana but given that I love consensus and coordination algorithms, I was initially pulled in by the use of a verifiable delay function as a coarse network clock within the PBFT implementation (broadly referred to as "Proof of History").
Venture Capital
At Moonfire, we believe that pre-seed and seed capital is underfunded in Europe. Europe captures less than 20% of global VC capital despite representing over 25% of GDP. On top of this, the rate of European Series A investment is increasing faster year over year. At Moonfire, we're helping proliferate a strong pre-seed and seed ecosystem of breakthrough entrepreneurs using software, data, and machine learning.
Machine Learning for sourcing, screening, and evaluation
- At Moonfire, I spend most of my time building systems which aim to use scalable data infrastructure and learned models to reason about all aspects of the venture capital lifecycle.
Venture fund modelling, forecasting, and simulation
- Traditional VC fund models are often created at the beginning of a fund and not maintained. This makes it challenging to be confident adapting your strategy to evolving market conditions. I'm passionate about using statistics, simulations, and projections to maintain realtime insight into the diversity of potential futures.
Web3 venture capital
- As angel syndicates grow, trust between investors diminishes quickly. While crypto-native investment DAOs have established incredible precedent, I'm working on managing traditionally regulated venture capital firms and angel syndicates from the ground up with web3 principles.
Distributed Systems
I've spent a significant majority of my career working on large, stateful, distributed systems in high-throughput production environments. Within the stack, I'm really passionate about coordination and communication protocols, performance (of both systems and engineers), and autonomous infrastructure operations.
Kubernetes
- I’ve been using Kubernetes in production since 2016. I was also a member of the Kubernetes core contributor team from 2018 until 2021 where I served on the Release Team for four releases and participated in several Working Groups and Special Interest Groups.
Search and recommendation
- Two of my endurring passions have long been deep learning for natural language processing and distributed systems. Modern search engines and recommender systems are an awesome intersection of these two worlds.
Consensus and coordination
- I am passionate about the theory and practice of building systems which need to maintain a coordinated consensus to manage the distribution of actions and information.