Edward De Brouwer

Research Scientist at Genentech

[Edward De Brouwer @EdwardOnBrew]

Very cool work by our intern @namkye0n9 at Genentech last fall, exploring how agentic systems can accelerate drug discovery. 👇🤖💊

[Namkyeong Lee @namkye0n9]

🤔 Can general-purpose LLMs tackle domain-specific challenges like drug discovery?
Our answer is yes, when they collaborate as multi-agents!

Our project, conducted during my internship at Genentech, is now available in arXiv!

Paper: https://t.co/fvFSc2ennc

[Edward De Brouwer @EdwardOnBrew]

More info in our paper :

https://t.co/NA6SL1j7bN

Credits : @rahulgk, @VectorInst, @UofT

[Edward De Brouwer @EdwardOnBrew]

Already the last day at @iclr_conf where we presented our work with @rahulgk on

Augmenting Neural ODEs with long range memory using orthogonal projection dynamics 🚀

https://t.co/NA6SL1j7bN

[Edward De Brouwer @EdwardOnBrew]

Here is an example of the quality of reconstruction of our model for clinical time series 😍 After processing the whole time series, we aim at reconstructing the past from hour=24.

[_hylandSL - not here @_hylandSL]

Self-promotion/related work: last summer my intern @EdwardOnBrew worked on models for estimating time-varying treatment effects: https://t.co/Br3Qtxsqaw

[Bastian Grossenbacher-Rieck @Pseudomanifold]

Really proud how this one turned out! 😊

We provide a framework for developing general curvature measures for hypergraphs, analysing their respective strengths and weaknesses.

Stay tuned for more information!

[Edward De Brouwer @EdwardOnBrew]

Ultimately, modeling uncertainties in the individual treatment effects predictions can potentially lead to more informed treatment decision and better deferral strategies, as shown in our experiments.

[Edward De Brouwer @EdwardOnBrew]

Predicting the effect of treatments on patients with *uncertainty* 📈
My talk at @FieldsInstitute is available here : https://t.co/F1AoI4Zrwu. Joint work with @_hylandSL @javiergonzh

[Edward De Brouwer @EdwardOnBrew]

Paper is available https://t.co/OHK4NAObmL
In this paper we relax the limiting overlap assumption by explicitely modeling the lack of support for individual treatment effect estimation through epistemic uncertainty (as initially suggested by @anndvision, @yaringal and others)

[Edward De Brouwer @EdwardOnBrew]

We model epistemic uncertainties by learning a Neural Stochastic Differential Equation that parametrizes the posterior of the hidden temporal process, leading to a varitional posterior for neural differential equations.

[Edward De Brouwer @EdwardOnBrew]

Excited to share my new work : "Deep Counterfactual Estimation with Categorical Background Variables" to appear at #neurips 🎉 This work tries to relax the strong assumptions needed for counterfactual reconstruction by allowing for a controlled error. https://t.co/MszkSe0Jb3

[Bastian Grossenbacher-Rieck @Pseudomanifold]

@EdwardOnBrew talking about "Topological Graph Neural Networks".

📜 https://t.co/tpl43yIGiM

🖥️https://t.co/OV98aqrkbW

#topology #MachineLearning

[Edward De Brouwer @EdwardOnBrew]

@MaxIlse @eric_nalisnick Because you can represent them as NODE normalizing flows, they might suffer from the same issues ? -> https://t.co/Yshka6yiML

[Edward De Brouwer @EdwardOnBrew]

Another passionate discussion about #topology and #GNN ! Thanks to @HannesStaerk for the invitation :) @Pseudomanifold @ExpectationMax @Michael_D_Moor @kmborgwardt

[Edward De Brouwer @EdwardOnBrew]

@Michael_D_Moor @Stanford it looks amazing ! Love the reference to @3blue1brown topological video <3

[Hannes Stark @HannesStaerk]

Today in 1.4h in the GraphML reading group @EdwardOnBrew presents his paper "Topological Graph Networks"! https://t.co/bE6qynzH0q

Join on Zoom at 3pm UTC! https://t.co/9fxz75bW1o

Co-authors:
@ExpectationMax @Michael_D_Moor Yves Moreau @Pseudomanifold @kmborgwardt

[Edward De Brouwer @EdwardOnBrew]

Super excited to spend the next 4 months with @rahulgk to advance machine learning for healthcare at @UofT and @VectorInst 🚀🚀🚀