naresh_xai

I think it can be done. Just maintain the data not in a standard tabular format but a time based graph instead. And put constraints to not look back. Constrained deep learning is a thing too.

Current set of tools in explainable AI do allow deep neural nets to tell why they gave the response that they did to a certain extent. Contact me if you want to see it applied to text based DL models, time series based DL models or Image based DL models.

Current set of tools in explainable AI do allow deep neural nets to tell why they gave the response that they did to a certain extent.

Contact me if you want to see it applied to text based DL models, time series based DL models or Image based DL models.

Are we still limiting to visual cues and not the auditory,smell,taste,touch data which we get exposed to?

Umm, no there are clear verification methods for Explainable AI techniques today. One way to check the justification would be if things which were important in the justification were removed in some sense, then would the output change signficantly. Sort of like a sensitivity test for justification.

That’s basically the gist of explainable AI

Pretty cool stuff

If revenue was more in a non-Apple ecosystem, would it not be the case that devs would all move to a non-Apple ecosystem.

Android also has the problem that users are far less likely to be as educated (since target audience is not white collar but blue collar) and far less likely to report problems, issues or malware. So I'd even take the reported values to be lower than what the actual case is.

But apple does not allow for that. So in that sense, they do not allow businesses for targeted advertisement. Please correct me if I am wrong.

Their processes and org structure is not built for that either with siloed teams. So much that quite often some teams will not know what kind of data is being collected by other teams even.

And 2 weeks later, the guy will resell stuff from another account from different paid reviews. Which are also always evolving to capture consumer trust. Works especially well for products in usd 1k range.

Companies list/re-list products on amazon faster than you can say ‘go’ unfortunately.

The number of loans a bank gives out to such a specific factor might be low enough that it would take 30 years for a bank to have an actual study. That's 1-2 generations of abuse.

Most banks just copy each other.

Unless you have causal proof, its irresponsible for a business to use such factors in modeling outcomes.

I also think that it is possible that the model learned that information from too small of a data sample. What is a good data sample for every such feature in a relatively balanced manner is really difficult to build a dataset from.

Consider a sample size of 10/1million with height value of 7m. And somehow 7/10 had poor ability to repay loans. With such a small sample size of this relevant factor be a good thing to rely on?

AI with no dibiasing does not work often enough. I’ve seen enough examples in Computer Vision models to say so. (Even segmentation models tend to rely on external clues and this affects the generalizability of the model). Happy to share more via email.

Happy to share a lot more ethical and challenges of non transparent AI via email with you. Bigger companies are pretty much hiding behind non transparent models and trying to ignore their failures at the moment.

Exactly why we need causal reasoning/causal proof alongwith ML

Interested in seeing where symbolic AI crowd has disagreed with that. Only group of people who disagree that I know of are a small set of people who think you should build inherently explainable models as opposed to explaining the decisions of deep neural networks.

Happy to share decks and references to your email address. Please share your email address with me :)

Graph Neural networks are currently used a lot in the neural networks for drug discovery space. They significantly beat RNN and CNN baseline/complex versions equivalents on the same datasets(Tox21, QM9 efc).

@Paul Robinson: Did you look at the number of mislabeled images in the udacity self driving car dataset? If you do not have an understanding of the subject yourself, you’re likely to make the same annotation errors which feed into a bad model.

Think more in terms of function decomposition rather than having to look at each individual parameter and you will find papers and techniques which lead to deep neural networks being quite explainable.

Contrastive LRP would be a good starting point for you for generating high fidelity explanations at any point in the network.

Again the non-sequitur argument of an explainable model must be worse than a deep learning system and there has to be a tradeoff.

You don’t need to reduce complexity to induce explainability. You just need to decompose the function into smaller parts which you can understand.

Contrastive LRP for example is a Function decomposition technique for explaining deep neural networks with high fidelity.

You should definitely look at advances in the field then. There a lot more promising work beyond grad cam. And there are a few techniques for human oriented explanations - namely TCAV (human oriented explanations) and PatternNet/PatternAttribution as well as contrastive LRP (in terms of much more informative heatmapping techniques which are dependent on intermediate layers as well.)

Not to mention the fact that most models are trained without a background class and tend to give overconfident predictions on out of distribution samples.

Let’s see. Robotics problems were claimed to be tractable through AI. However, a large majority of robotics solutions today are 90% derived through control systems (which follow some degree of causal analysis) followed by AI to optimize the last few percentages if possible.

Finding something odd for an algorithm (especially a deep neural network) is hard because they fail in just so many ways. For example, lenet for mnist almost always gives high confidence predictions for random tensors(torch.randn). Most imagenet models fail in the presence of just 20-30% salt and pepper noise. (Both of these are problems solvable through simple preprocessing techniques)

Or you can actually visualize and quantify impact of aggregates of meaningful features within networks using the methodology described in NetDissect or in TCAV. But of course, that casts doubt into a lot of claimed mechanisms and tons of ML research as I have found out.

If you were ever involved in the drug discovery process, then you would know that statistical evidence through clinical trials is only for the last ounce of drug testing. Out of 10,000+ drug molecule candidates only ~ 5 molecules get selected for clinical trials. A lot of molecules are removed as candidates for toxic components/ solubility issues/ chemical stability issues etc.

So even when we do not understand the precise impact of drugs on humans and there is no safer mechanism to test, we leave only 0.5% of candidate molecules to empirical/statistical evidence in the form of clinical trials.

On the other hand if drug discovery was treated as a pure AI problem, we would have thousands of unverified and unsafe molecules in clinical trials.

Causal principles get us to 99.5% of the way in drug discovery. Unfortunately not so in AI.

Which is strongly misleading tradeoff. For a ton of tasks, deep learning methods are no better than white box regression or tree ensemble methods.

And there is no reason to expect that a deep learning model has to be unexplainable. He’s putting up a mystical interpretability vs accuracy tradeoff which does not exist.

There is a ton of research out there making deep neural networks (slightly to significantly interpretable)

When a person’s claim to fame and research is dependent on ignoring explainability and causality in research, he will ignore it to the best of his means.

To him, all the precursors to clinical trials (selection of a molecule from restricted set of molecules which satisfy certain causal criteria) then followed by extensive multi million dollar experiments to safeguard the last 1-2% of uncertainty is equivalent to a barely preprocessed neural network.

I mean let’s ignore the basic principles of pharmacology and medicine and just run every possible molecule through humans since that is a valid approach according to him.