Kerry Champion (@MenloML) header
Kerry Champion (@MenloML) profile picture

Let's make AI more trustworthy, verifiable and explainable.

Palo Alto, CA
Joined December 2017
Photo by Kerry Champion (@MenloML)Photo by Kerry Champion (@MenloML)Photo by Kerry Champion (@MenloML)Photo by Kerry Champion (@MenloML)Photo by Kerry Champion (@MenloML)Photo by Kerry Champion (@MenloML)Photo by Kerry Champion (@MenloML)Photo by Kerry Champion (@MenloML)Photo by Kerry Champion (@MenloML)Photo by Kerry Champion (@MenloML)
In statistics, a small sample often gives a weird result, especially if the samples aren't independent. A family is a small, non-independent sample of the population. This is why your family is so weird.
13
145
27
658
Kerry Champion retweeted
A deer must have lost its keys. ⁦@EconTalker⁩
4
1
28
"Interpretability slows down ML adoption" Model-agnostic tools also invalidate this argument. It is also contraire to my experiences from jobs, consulting and dozens of conversations with practitioners and statisticians: LACK of interpretability is slowing down adoption.
2
2
1
18
Performance > Interpretability True, if you can capture 100% of your problem's definition in a single loss function, the training data was collected in controlled setting, and you do not care to learn anything about the problem. An unlikely situation.
3
3
1
32
Most common arguments against interpretable ML: - Humans can't explain their actions either - Performance > Interpretability - Slows down ML adoption - Linear model also not interpretable - Might create illusion of understanding the model Thread:
10
139
24
323
Kerry Champion retweeted
The #ExplainableAI wave reaches France (think I remember an AINow report advised this?) cc @katecrawford @FrankPasquale "Any algorithm that can't be explained can't be used by the civil service" promises France's digital minister Mounir Mahjoubi in French parliament [RGPD = GDPR] xitter.gq/gchampeau/status/96135…
"Aucun algorithme non explicable ne pourra être utilisé par l'administration"; et aucun algorithme en matière de condamnations judiciaires, promet @mounir. #RGPD #DirectAN
3
19
3
34
Kerry Champion retweeted
I think that's the key: we don't often know the right question to ask when we start a line of empirical inquiry. We refine some initial vague set of question as we interact with data, answer some of them, and generate new ones.
3
13
3
105
Kerry Champion retweeted
I'm as enthusiastic about the future of AI as (almost) anyone, but I would estimate I've created 1000X more value from careful manual analysis of a few high quality data sets than I have from all the fancy ML models I've trained combined.
18
299
50
1,027
More #ExplainableAI and #FATML work from @TheEconomist. Nice summary of @DARPA #XAI program and performers. I feel like every major pub is putting out their "#AI is not going to save us" article. Kind of fun. economist.com/news/science-a… via
3
2
One of my main concerns about machine learning interpretability tools is that they will make people think they understand ML when they don't. People seem to think linear models are interpretable, but no one looks at them and has the intuition that they have adversarial examples xitter.gq/superwuster/status/963…
17
210
17
717
2nd release this week: version 0.2 of the interpretable ML book 💪 christophm.github.io/interpr… New chapters: - shapley value - on explanations - short stories Reworked: - RuleFit - interpretability And I could use some help drawing cats: https://xitter.gq/t.co/xWlSuGuEOc 🐱
3
14
1
40
Kerry Champion retweeted
I recently spoke with @hugobowne for DataFramed, the @DataCamp podcast, about #datascience and online experiments at @Etsy. We had a great time! Check it out here: https://xitter.gq/t.co/L5PQ2KPGc9
5
17
2
58
No more excuses for opaque machine learning models! iml, the Interpretable Machine Learning package for R is out! 🎉🎉🎉 Featuring partial dependence plots, feature importance, LIME and more. Available on github and - soon - your favorite CRAN mirror! github.com/christophM/iml
1
64
4
163
Kerry Champion retweeted
Optimist: AI has achieved human-level performance! Realist: “AI” is a collection of brittle hacks that, under very specific circumstances, mimic the surface appearance of intelligence. Pessimist: AI has achieved human-level performance.
73
3,382
141
9,121
"Debugging is like being the detective in a crime drama where you are also the murderer."
51
1,078
63
3,008
This is why we need #ExplainableAI #xai #fatml Is your software racist? politi.co/2EMgkhZ
1
1
Causal Learning and Explanation of Deep Neural Networks via Autoencoded Activations. arxiv.org/abs/1802.00541
1
6
Kerry Champion retweeted
"we will soon see the creation of a category of human professionals who don’t have to make the moment-to-moment decisions themselves but instead manage a team of A.I. workers". By @a16z's @vijaypande nyti.ms/2GeMiDL
4
4
Kerry Champion retweeted
Reading research on adversarial examples is like watching an alien species learn about optical illusions, and that's pretty cool
4
12
1
87
Kerry Champion retweeted
A nice report of Kaggle's activities in 2017 (Google acquired Kaggle last year). Lots of datasets and contests, lots more machine learning and data scientist users, lots of usage for academic classes. I take solace that the #1 downloaded dataset was the "World Happiness Report"! xitter.gq/kaggle/status/95556579…
Read our CEO @antgoldbloom's detailed recap of Kaggle in 2017 and our big plans for 2018 goo.gl/BzzK6y
1
45
190