@vahidbalazadeh.bsky.social
Worried about reliability?
CausalPFN has a built-in calibration, and can make reliable estimations even for datasets that fall outside of its pretraining prior.
Try it using: pip install causalpfn
Made with ❤️ for better causal inference
[7/7]
#CausalInference #ICML2025
CausalPFN has a built-in calibration, and can make reliable estimations even for datasets that fall outside of its pretraining prior.
Try it using: pip install causalpfn
Made with ❤️ for better causal inference
[7/7]
#CausalInference #ICML2025
June 11, 2025 at 1:13 PM
Worried about reliability?
CausalPFN has a built-in calibration, and can make reliable estimations even for datasets that fall outside of its pretraining prior.
Try it using: pip install causalpfn
Made with ❤️ for better causal inference
[7/7]
#CausalInference #ICML2025
CausalPFN has a built-in calibration, and can make reliable estimations even for datasets that fall outside of its pretraining prior.
Try it using: pip install causalpfn
Made with ❤️ for better causal inference
[7/7]
#CausalInference #ICML2025
When does it work?
Our theory shows that posterior distribution of causal effects is consistent if and only if the pretraining data only includes identifiable causal structures.
👉 We show how to carefully design the prior, one of the key differences in our work relative to predictive PFNs. [6/7]
Our theory shows that posterior distribution of causal effects is consistent if and only if the pretraining data only includes identifiable causal structures.
👉 We show how to carefully design the prior, one of the key differences in our work relative to predictive PFNs. [6/7]
June 11, 2025 at 1:13 PM
When does it work?
Our theory shows that posterior distribution of causal effects is consistent if and only if the pretraining data only includes identifiable causal structures.
👉 We show how to carefully design the prior, one of the key differences in our work relative to predictive PFNs. [6/7]
Our theory shows that posterior distribution of causal effects is consistent if and only if the pretraining data only includes identifiable causal structures.
👉 We show how to carefully design the prior, one of the key differences in our work relative to predictive PFNs. [6/7]
Real-world uplift modelling:
CausalPFN works out of the box on real-world data. On 5 real RCTs in marketing (Hillstrom, Criteo, Lenta, etc.), it outperforms baselines like X-/S-/DA-Learners on policy evaluation (Qini score). [5/7]
CausalPFN works out of the box on real-world data. On 5 real RCTs in marketing (Hillstrom, Criteo, Lenta, etc.), it outperforms baselines like X-/S-/DA-Learners on policy evaluation (Qini score). [5/7]
June 11, 2025 at 1:13 PM
Real-world uplift modelling:
CausalPFN works out of the box on real-world data. On 5 real RCTs in marketing (Hillstrom, Criteo, Lenta, etc.), it outperforms baselines like X-/S-/DA-Learners on policy evaluation (Qini score). [5/7]
CausalPFN works out of the box on real-world data. On 5 real RCTs in marketing (Hillstrom, Criteo, Lenta, etc.), it outperforms baselines like X-/S-/DA-Learners on policy evaluation (Qini score). [5/7]
Benchmarks:
On IHDP, ACIC, Lalonde:
– Best avg. rank across many tasks
– Faster than all baselines
– No tuning needed compared to the baselines (that were tuned via cross-validation)
[4/7]
On IHDP, ACIC, Lalonde:
– Best avg. rank across many tasks
– Faster than all baselines
– No tuning needed compared to the baselines (that were tuned via cross-validation)
[4/7]
June 11, 2025 at 1:13 PM
Benchmarks:
On IHDP, ACIC, Lalonde:
– Best avg. rank across many tasks
– Faster than all baselines
– No tuning needed compared to the baselines (that were tuned via cross-validation)
[4/7]
On IHDP, ACIC, Lalonde:
– Best avg. rank across many tasks
– Faster than all baselines
– No tuning needed compared to the baselines (that were tuned via cross-validation)
[4/7]
Why does it matter?
Causal inference traditionally needs domain expertise + hyperparameter tuning across dozens of estimators. CausalPFN flips this paradigm: we pay the cost once (at pretraining), then it’s ready to use out-of-the-box! [3/7]
Causal inference traditionally needs domain expertise + hyperparameter tuning across dozens of estimators. CausalPFN flips this paradigm: we pay the cost once (at pretraining), then it’s ready to use out-of-the-box! [3/7]
June 11, 2025 at 1:13 PM
Why does it matter?
Causal inference traditionally needs domain expertise + hyperparameter tuning across dozens of estimators. CausalPFN flips this paradigm: we pay the cost once (at pretraining), then it’s ready to use out-of-the-box! [3/7]
Causal inference traditionally needs domain expertise + hyperparameter tuning across dozens of estimators. CausalPFN flips this paradigm: we pay the cost once (at pretraining), then it’s ready to use out-of-the-box! [3/7]
What is it?
CausalPFN transforms effect estimation to a supervised learning problem. It's a transformer trained on millions of simulated datasets. It learns to map from data to treatment effect distributions directly. At test time, no finetuning and manual estimator selection are required. [2/7]
CausalPFN transforms effect estimation to a supervised learning problem. It's a transformer trained on millions of simulated datasets. It learns to map from data to treatment effect distributions directly. At test time, no finetuning and manual estimator selection are required. [2/7]
June 11, 2025 at 1:13 PM
What is it?
CausalPFN transforms effect estimation to a supervised learning problem. It's a transformer trained on millions of simulated datasets. It learns to map from data to treatment effect distributions directly. At test time, no finetuning and manual estimator selection are required. [2/7]
CausalPFN transforms effect estimation to a supervised learning problem. It's a transformer trained on millions of simulated datasets. It learns to map from data to treatment effect distributions directly. At test time, no finetuning and manual estimator selection are required. [2/7]
Our general approach can be applied to various settings like bandits, MDPs, and POMDPs (5/5)
❤️ w/ Keertana Chidambaram, Viet Nguyen, @rahulgk.bsky.social , and Vasilis Syrgkanis
Link to paper: arxiv.org/abs/2404.07266
❤️ w/ Keertana Chidambaram, Viet Nguyen, @rahulgk.bsky.social , and Vasilis Syrgkanis
Link to paper: arxiv.org/abs/2404.07266
Sequential Decision Making with Expert Demonstrations under Unobserved Heterogeneity
We study the problem of online sequential decision-making given auxiliary demonstrations from experts who made their decisions based on unobserved contextual information. These demonstrations can be v...
arxiv.org
December 12, 2024 at 4:05 PM
Our general approach can be applied to various settings like bandits, MDPs, and POMDPs (5/5)
❤️ w/ Keertana Chidambaram, Viet Nguyen, @rahulgk.bsky.social , and Vasilis Syrgkanis
Link to paper: arxiv.org/abs/2404.07266
❤️ w/ Keertana Chidambaram, Viet Nguyen, @rahulgk.bsky.social , and Vasilis Syrgkanis
Link to paper: arxiv.org/abs/2404.07266
To do so, we consider all prior distributions on the unobserved factors (e.g. the distribution over each arm's mean reward) that align with the expert data. We then choose the prior with the maximum entropy (least information) and apply posterior sampling to guide the exploration (4/5)
December 12, 2024 at 4:05 PM
To do so, we consider all prior distributions on the unobserved factors (e.g. the distribution over each arm's mean reward) that align with the expert data. We then choose the prior with the maximum entropy (least information) and apply posterior sampling to guide the exploration (4/5)
Online exploration can eventually identify unobserved factors but requires trial and error. Instead, we use expert data to limit the exploration space. In a billion-armed bandit with expert data spanning only the first ten actions, the learner should only explore those ten arms (3/5)
December 12, 2024 at 4:05 PM
Online exploration can eventually identify unobserved factors but requires trial and error. Instead, we use expert data to limit the exploration space. In a billion-armed bandit with expert data spanning only the first ten actions, the learner should only explore those ten arms (3/5)
Unobserved confounding factors affect the expert policy in ways that are not understood by the learner. An important example is experts acting with privileged information. Naive imitation leads to single aggregated policies for each observed state and fails to generalize (2/5)
December 12, 2024 at 4:05 PM
Unobserved confounding factors affect the expert policy in ways that are not understood by the learner. An important example is experts acting with privileged information. Naive imitation leads to single aggregated policies for each observed state and fails to generalize (2/5)