Professor of Psychology. I use machine learning to investigate language semantics, reasoning, with application to the digital phenotyping of mental illness.
Emory University, Atlanta, GA
Joined June 2014
- Tweets63
- Following76
- Followers84
- Likes31
Artificial intelligence classifies primary progressive aphasia from connected speech pubmed.ncbi.nlm.nih.gov/3891…
Phillip Wolff retweeted
Just published in Brain!
We used LLMs to classify (primary progressive aphasia) PPA in a data-driven way.
We found verb frequency is a critical variable in PPA classification. Saying "negotiate" versus "talk" makes a difference!
academic.oup.com/brain/advan…
@DickersonLabMGH @PM_Wolff
Stacy Abrams @staceyabrams made Biden & Kamala's win in Georgia possible. What an amazing woman! It was a moral victory for the nation.
As Joe Biden took a narrow lead in Georgia on Friday, it was Stacey Abrams who was celebrated, a sign of her remarkable ascent as a power broker since her failed bid for governor of that state two years ago. nyti.ms/36t8CHV
2
Phillip Wolff retweeted
Welcome back America! Congratulations to @JoeBiden and @KamalaHarris for their election! While we are about to celebrate the 5th anniversary of the Paris Agreement, this victory symbolizes our need to act together more than ever, in view of climate emergency. 🇺🇸
#Election2020
What a fantastic position! You never stop impressing me!
I'm thrilled to accept a data science role at @WSJ with @rossfadely, @JeffersTess and team focusing on building great digital products with data!
Special thanks to Matt Graci, Des Seraph, and @InsightFellows (Allen Chang, @eekearney, Kim Vo and my wonderful cohort of fellows)
1
7
Phillip Wolff retweeted
When conservatives attack their enemies, they do so from a position of social inferiority, so their attacks are resentful. When left attacks, it does so from a position of perceived moral superiority, so its attacks are filled with ridicule and scorn. nytimes.com/2019/12/12/opini…
Phillip Wolff retweeted
.@SuryaGanguli & I are co-organizing the next @StanfordHAI Conference Apr 1, 2020—no joke!—on the topic Triangulating Intelligence: Melding Neuroscience, Psychology, and AI. Botvinick—@YejinChoinka—@chelseabfinn—@AudeOliva—Tenenbaum—@dyamins—save the date! hai.stanford.edu/events/tria…
A terrific article identifying the strengths and limitations of the language model BERT.
New paper up! “What BERT is not: Lessons from a new suite of psycholinguistic diagnostics for language models” arxiv.org/abs/1907.13528. Tests that have elicited informative patterns in the brain prove quite useful for clarifying strengths and limitations of BERT pre-training.
1
Emory Psyc Department is conducting searches for positions in Human Cognition (apply.interfolio.com/65505) and Developmental Science (apply.interfolio.com/64756). Those using AI, Machine Learning and/or NLP methods to study cognition are especially encouraged to apply.
Phillip Wolff retweeted
"The model uses 8.3 billion parameters and is 24 times larger than BERT and 5 times larger than OpenAI’s GPT-2"
Why am I not surprised? For such a computational effort, I hope the weights will be released publicly...
venturebeat.com/2019/08/13/n…
This is an outstanding paper (biorxiv.org/content/10.1101/…) with results that are highly compatible with the language-specific predictive processes instantiated in transformer networks like BERT and XLNet.
Very exited about this work led by Cory Shain and @IdanAsherBlank, also with @marty_with_an_e and William Schuler:
biorxiv.org/content/10.1101/…
" fMRI reveals language-specific predictive coding during naturalistic sentence comprehension" 1/4
Phillip Wolff retweeted
One of my favorite ideas from #cog2019 #cogsci2019 so far . You can build human heuristics into machine learning models using something as simple as regularization. From Özgür Şimşek
Cool study showing how humans are able to identify objects from nearly any angle. disq.us/t/3g1vbz1
3
Artificial intelligence could spot early signs of schizophrenia newscientist.com/article/220…
Phillip Wolff retweeted
XLNet: a new pretraining method for NLP that significantly improves upon BERT on 20 tasks (e.g., SQuAD, GLUE, RACE)
arxiv: arxiv.org/abs/1906.08237
github (code + pretrained models): github.com/zihangdai/xlnet
with Zhilin Yang, @ZihangDai, Yiming Yang, Jaime Carbonell, @rsalakhu
Phillip Wolff retweeted
Those guys at Boston Dynamics are continuing to break the boundaries
@XClaimItNews are first to the news that matters to you
#ArtificialIntelligence #XClaim #InnovatingCreativity
