Filipe Rodrigues

21 posts

Filipe Rodrigues banner
Filipe Rodrigues

Filipe Rodrigues

@fmpr87

Associate Professor at Denmark Technical University (DTU)

Denmark Se unió Ocak 2016
81 Siguiendo51 Seguidores
Filipe Rodrigues retuiteado
Daniele Gammelli
Daniele Gammelli@DanieleGammelli·
Predictions from ML models are often used for downstream decision-making tasks, where a sequential predict-then-optimize strategy is used to train the ML model. But do accurate predictions translate into better decisions? Spoiler: NO arXiv: arxiv.org/abs/2108.00858 🧵(1/5)
English
2
1
5
0
Filipe Rodrigues
Filipe Rodrigues@fmpr87·
Is a complex deep neural network better than a simple linear regression for spatio-temporal forecasting in transportation problems? Check here how a simple linear regression can provide comparable results to many state-of-the-art approaches: arxiv.org/abs/2203.02954
English
3
2
14
0
Filipe Rodrigues retuiteado
Daniele Gammelli
Daniele Gammelli@DanieleGammelli·
Join us at #INNF Workshop #ICML2021 at 11am UTC (9am CET) today! The efficiency of mobility systems depends on the ability to model the need for transportation. We show how LVMs and normalizing flows enable us to generate distributions matching complex urban topologies.
Daniele Gammelli tweet media
English
1
1
1
0
Filipe Rodrigues retuiteado
Daniele Gammelli
Daniele Gammelli@DanieleGammelli·
How can we learn *transferrable* control policies for Autonomous Mobility-on-Demand systems across cities? Our recent work shows how graph neural networks and deep RL agents exhibit promising zero-shot transfer capabilities. Up on arXiv! arxiv.org/abs/2104.11434
Daniele Gammelli tweet media
English
1
5
12
0
Filipe Rodrigues
Filipe Rodrigues@fmpr87·
Cool work from our PhD student @DanieleGammelli on graph neural network reinforcement learning for autonomous mobility-on-demand systems in collaboration with @StanfordASL!
Daniele Gammelli@DanieleGammelli

How can we learn *transferrable* control policies for Autonomous Mobility-on-Demand systems across cities? Our recent work shows how graph neural networks and deep RL agents exhibit promising zero-shot transfer capabilities. Up on arXiv! arxiv.org/abs/2104.11434

English
0
0
3
0
Filipe Rodrigues
Filipe Rodrigues@fmpr87·
Check out our new AAAI 2018 paper: "Deep Learning from Crowds" (arxiv.org/abs/1709.01779). Something to consider the next time you train a deep NN on crowdsourced labels...
English
0
0
1
0

Descubrir

@DanieleGammelli @mlsm_dtu @StanfordASL @inonpeled @itssieee @dtumanagement @MarkouIoulia @elonmusk