Janis Klaise

Machine Learning @seldon_io | Developer of explainability library Alibi Explain and outlier & drift detection library Alibi Detect

[Janis Klaise @JKlaise]

Very pleased to release Alibi with @arnaudvl3 and @seldon_io - an open source library for #MachineLearning model inspection and interpretation. Currently includes Anchors (@guestrin), contrastive explanations (@pinyuchenTW) and trust scores (@_beenkim): https://t.co/DS2xMyXifY

[Janis Klaise @JKlaise]

@fishnets88 I found the same problem on stack exchange and it reduces to finding integer points on a hyperbolic curve for which there is a known method to generate all such points! https://t.co/hAZyG2UJf1

[Janis Klaise @JKlaise]

@fishnets88 Luckily WolframAlpha gives the first few solutions: https://t.co/xw8aCBnBEB (the negative branch doesn't give any extra):
r=3, b=1 (3/4*2/3=0.5)
r=15, b=6 (15/21*14/20=0.5)
r=85, b=35 (85/120*84/119=0.5)

[Janis Klaise @JKlaise]

Awesome work by the team! Kernel based drift detection accelerated using KeOps!

[Seldon Research @SeldonResearch]

We are excited to announce the release of Alibi Detect v0.11.0, featuring widened serialisation support and a new backend that allows drift detection to be rapidly performed on large datasets. https://t.co/q8lp8MaZZY

[Janis Klaise @JKlaise]

Great work from the team with a special shout-out to @RSamoilescu !

[Seldon Research @SeldonResearch]

We are pleased to announce the release of Alibi Explain v0.9.0 with support for calculating global feature importance via Permutation Importance or Partial Dependence Variance. https://t.co/W9kNoM0D4c

[Janis Klaise @JKlaise]

New version of Alibi Explain out with a focus on explaining the global behaviour of any black-box model via partial dependence plots! Shout-out to @RSamoilescu for the contribution and to the rest of the team @seldon_io for numerous improvements!

[Seldon Research @SeldonResearch]

We are delighted to announce the release of Alibi Explain https://t.co/W9kNoMhG6c v0.8.0 featuring support for Partial Dependence plots, enabling global explainability of any model.

[Janis Klaise @JKlaise]

Nice work by the DS team @seldon_io! Special thanks to @scillitoes and @oblibob for making Alibi easier to use with every release!

[Seldon Research @SeldonResearch]

Monitoring for drift can be a complex endeavour, involving multiple ML components. To make life easier, we are pleased to announce the release of Alibi Detect v0.10.0!

[Janis Klaise @JKlaise]

Check out one of the newest additions to Alibi Explain - a prototype selection method that can help you understand your datasets better. Shout-out to @RSamoilescu for leading the work!

[Seldon Research @SeldonResearch]

We are happy to introduce a new XAI tool in Alibi Explain 0.7.0 arsenal. ProtoSelect - Bien and Tibshirani (2012) https://t.co/FXzsxSN8nq - is a prototype selection method to construct a condensed view of a dataset and an interpretable classifier: https://t.co/XMh2RXiv5p

[Janis Klaise @JKlaise]

Check out some details on the new similarity explanation method available in Alibi Explain 0.7.0!

[Seldon Research @SeldonResearch]

The recent Alibi Explain 0.7.0 release includes a new class of explainers. Similarity-based explanations are probably one of the most intuitive and visual explanations one can get: https://t.co/ieJZ6p5X1U

[Janis Klaise @JKlaise]

Very pleased to have our paper on context-aware drift accepted at #ICML2022! See the thread below for motivation and overview of the method 👇 @seldon_io @SeldonResearch

[Seldon Research @SeldonResearch]

Thrilled to have our work on context-aware drift detection accepted to #ICML2022. We look forward to discussing it with the community!

[Janis Klaise @JKlaise]

Happy to share the newest open-source work from the data science team @SeldonResearch @seldon_io !

[Seldon Research @SeldonResearch]

Announcing Alibi Explain 0.7.0 with a new class of similarity-based explainers and data sumarization techniques: https://t.co/nXa4Hz8MYe

[Janis Klaise @JKlaise]

Would anyone like some Fibonacci soup I made? https://t.co/HAZKuSjpqi

[Janis Klaise @JKlaise]

Proud to see @seldon_io represented at @aistats_conf with original research on calibrated multivariate change detection! Shoutout to @ollyjcobb for spearheading the research!

[Seldon Research @SeldonResearch]

Looking forward to discussing this work at AISTATS tomorrow in the 0900-1030 BST poster session. Hope to see some of you there!

[Janis Klaise @JKlaise]

My friends: Fibonacci soup isn't real
Me: https://t.co/HAZKuSB0OS

[Janis Klaise @JKlaise]

Great new work by @ollyjcobb @arnaudvl3 and @scillitoes pushing our applied research in new and exciting directions!

[Seldon Research @SeldonResearch]

We're delighted to announce new research affording ML practitioners control over which changes to their data distribution should - and should not - be identified as drift.

Available now in Alibi Detect 0.9.0!

Docs: https://t.co/TC9nGn86Zd
Paper: https://t.co/cZFEXuDMti

🧵

[Janis Klaise @JKlaise]

@AsteroidTracker thanks for providing the cool NEOWS API (https://t.co/Pe29Gobot1)! I was wondering where the "is_potentially_hazardous" field comes from? I followed the data sources on the CNEOS website but didn't find this information there either.

[Seldon Research @SeldonResearch]

Nice to see our earlier findings in Alibi Detect for drift detection on graphs confirmed by this ICLR’22 paper! As shown in https://t.co/vgSxMB04un we also saw strong MMD drift detection performance using embeddings from a random GIN model on molecules with different scaffolds.

[Janis Klaise @JKlaise]

Massive shoutout to @oblibob who's been working on this for the last few months since joining @seldon_io! What a way to get immersed in a completely new topic!

[Janis Klaise @JKlaise]

A lot of new material here for anyone interested in the landscape of explainable machine learning!

[Seldon Research @SeldonResearch]

We're very excited to share our gentle yet comprehensive overview of explainability as part of the Alibi Explain documentation. We hope to help practitioners and anyone interested in XAI to orient themselves in this complex yet increasingly relevant topic https://t.co/OfWdRPo2jY.