Lasse Hansen: lassehansen.me

roberta rocca: https://rbroc.github.io/

Are these results for combining audio and text?

Thank you!

Are these prior results using the same dataset?

(I have lost my voice with laryngitis so I can’t unmute and interrupt…)

(sorry to hear!)

Thank you!

Wait, actually, I meant the “prior work” slide with results previously reported by other groups? Maybe three slides ago?

Yes!

Okay, got it!

I think it’s this meta-analysis from their group: https://pubmed.ncbi.nlm.nih.gov/31839552/

and this one for ASD also from their group: https://onlinelibrary.wiley.com/doi/pdf/10.1002/aur.1678?casa_token=c5cmcmmM7eUAAAAA:puxXjIgyciQ590Vu-W_Om2Osk6mMjrjM7OrqXv0_LcoLZ5c7qxsJSpa24DenfsQDCxd2FYXWLXjhmw

Thanks, Daniel!

Don't talk!

:)

What sort of diagnostic info do you have for the ASD group? Is it ADOS?

Do you have info about their language abilities, which are differentially impact on language? Like do yuo know if they have a language impairment as well?

It has a lot of typos!

Thanks!

your next satra!

🙂

I had a similar question about evaluation. Have you considered “hold one out” cross validation? For autism, I find this really helpful since they are such a heterogenous group, but of course, you are exploring a lot of different models.

Thank you for the great talk!

Thank you!

We've been learning that k-fold nested cross-validation provides the best statistical power

our postdoc Hamzeh has been putting together a manuscript to compare nested CV with traditional train/test. This work also results in the need for more data samples.

Thank you for a great talk!

Nicholas Cummins has a great paper looking into symptom-level analysis of depression: https://arxiv.org/pdf/2204.00088

Congrats!