I’m thrilled to share that I’ll be joining
@UCSDPsychology
as an Assistant Professor starting in July 2024! 🎉✨
This job is truly a dream come true. I feel beyond lucky to join such a wonderful department and to have the chance to build my own lab. (1/3)
Interested in how drawings of object categories change throughout childhood? Come by Friday morning at 11:50am at
#CogSci2018
to the early cognition session (Room MNQR)!
@judyefan
@mcxfrank
paper:
New paper out today 🎉in Behavioral Research Methods:
“The BabyView camera: Designing a new head‑mounted camera to capture children’s early social and visual environments.”
Example video clip below! And a short 🧵(1/11)
How might changes in children’s drawings reflect changes in their visual concepts? For our
#CogSci2019
paper (with
@judyefan
@mcxfrank
, ) we collected a large-scale database of children’s drawings (~13K) and analyzed changes in their visual features. (1/6)
📣 Excited to share that our paper, “Parallel developmental changes in children’s production and recognition of line drawings of visual concepts” has been published in Nature Communications! 🎉
📣I'm hiring a lab coordinator! This position is ideal for someone looking for research experience before applying to graduate school.
Deadline is April 1st—flexible summer start date. Details and instructions below!
📣 🎉Excited to announce that I'm recruiting graduate students and a postdoc to launch the Visual Learning Lab
@UCSDPsychology
in Fall 2024!
More information at —feel free to reach out with questions!
Excited to announce our new preprint, “Parallel developmental changes in children’s drawing and recognition of visual concepts” with Judy Fan (
@judyefan
), Zixian Chai, and Michael C. Frank (
@mcxfrank
): (1/13)
I'm thrilled to announce that next summer, I'll be starting as an Assistant Professor in the Department of Cognitive Science at
@UCSanDiego
.
I feel so fortunate to be joining such a wonderful department and research community. It's a dream I never dared to dream come true. (1/5)
Very excited to share that this paper with
@talia_konkle
, "Mid-level visual features underlie the high-level categorical organization of the ventral stream" is finally out! ($ wall).
Excited to announce a new paper: “Automated detections reveal the social information in the changing infant view” with Alessandro Sanchez, Allison M. Kraus,
@_ketan0
and
@mcfrank
at Child Development, ($) (1/15) 🧵
Excited to share a new preprint: “Developmental consistency in children’s drawings of object categories,” with Ying Wang, Stella Christie, Michael C. Frank
@mcxfrank
, & Judith E. Fan
@judyefan
. . A 🧵 (1/9)
Finally, if you find this kind of work interesting, I’m currently hiring!
I'm recruiting both a postdoc and a lab manager (more at ) to launch the Visual Learning Lab at
@UCSDPsychology
—please feel free to reach out (or come say hi at CDS)!
Excited to share that a new article with Mariko Moher, Susan Carey, and
@talia_konkle
is out at JEP:HPP ($wall, copy at ). We found evidence that preschoolers automatically process the real-world sizes of *depicted* objects (1/3)
Absolutely thrilled to announce that Jane Yang
@jane_yt_yang
will be an inaugural grad student in the Visual Learning Lab at
@UCSDPsychology
()!
I couldn't be more excited to have her join the lab and to get to do science together! Welcome Jane!!! 🎉🎉🎉
Excited to speak tomorrow at the
@svrhm2020
NeurIPS workshop (Dec 12th, 3pm PST/6pm EST) on my work with
@judyefan
, Zixian Chai, &
@mcxfrank
: "Parallel developmental changes in children's drawing and recognition of visual concepts." (1/8)
For now, if you’re struggling through the academic job market, I see you. This has been a long road, especially as an
#academicmama
of two.
Deeply grateful for the support of my family, friends, and mentors
@mcxfrank
@talia_konkle
@grez72
🙏 I wouldn’t be here without you! (3/3)
First Zoom sketchnotes! Fabulous talk by
@brialong
on what social cues are available within the infant’s view. Such a cool model and dataset. Turns out babies spend quite a lot of time looking at hands!
So grateful for our partnership with
@PurpleMuseum
! Check out their blog post on our work with
@judyefan
@mcxfrank
on children's drawings And if you're in the San Jose area, you can visit our drawing station and their wonderful children's museum!
My lab will examine how children “learn to see” using primarily empirical and computational methods (examples of recent work at ).
I’ll be hiring at all levels in the next year to start the lab with me in 2024—more to come! (2/3)
If you didn't have a chance to see our poster at CNS, here it is (along with a video walk through: ). If you want to talk about (perceptual) curiosity and how it might affect memory, let me know!
We are hiring! Come work with me 🙂
UC San Diego Psychology will be hiring two Assistant Professors this year in:
1. Social psychology.
2. Computational approaches to behavior and/or its neural mechanisms.
Apply by November 7th
As children got older, these category classifications became more accurate. And these gains in classification were not explainable by children’s ability to trace simple shapes or low-level effort covariates (amount of time spent drawing, number of strokes, ‘ink’ used). (4/6)
@brialong
’s
#CogSci2018
talk on how children’s drawings change through childhood: children’s drawings become more recognizable with age because they better capture relevant visual distinctions. Very cool work!
Many more details are in the paper and linked on the website: . We are currently collecting data and plan to share videos through
@databrary
!
Feel free to reach out with any questions! (11/11)
Taken together, these results suggest visual production and recognition of object categories improve throughout middle childhood. Preprint coming soon! (7/8)
On Mon 7/11 at 10AM PT: Our Dev Sci seminar featuring talks by
@Logan_Fiorella
("Motivation and metacognition in learning by drawing") &
@brialong
("Parallel developmental changes in children's production and recognition of drawings"), w/ broader discussion led by
@Moira_Dillon
!
Children were simply asked to touch the picture that was smaller *on the screen* on an iPad (same task/stim in Konkle & Oliva, 2012). But children were better at the task when the relative real-world sizes of the objects were congruent with their sizes in the real-world. (2/3)
I’ll be at
#CCN18
soon! Get in touch if you’re looking for a post doc on topics in high level visual representation using neuroimaging and modeling approaches.
Furthermore, this effect didn't vary according to how well children could identify the depicted objects—instead, adults & kids showed similar item effects. While a first step, these data hint that similar mechanisms underly size processing in adults & preschoolers! (3/3)
Do we continue to learn about which features best distinguish between categories (e.g. ‘dogs’ vs. ‘rabbits’) across childhood? We asked if children improve at (1) emphasizing these features when drawing and (2) using these features when recognizing drawings. (2/8)
@melissaekline
Thanks for talking about this openly! I also actively think about alternative career paths and encourage others to do the same. I’d love to land a faculty job, but the reality is that there *really* aren’t enough jobs for the number of qualified postdocs!
My lab, the Stanford Language and Cognition Lab (), is looking for a new Research Coordinator to start this summer. Lots of ways to get involved in research and learn new skills/methods in a fun, supportive environment!
We installed a free-standing station at the Children’s Discovery Museum in San Jose (
@PurpleMuseum
), where we collected over 37,000 drawings from children 2-10 years of age, and also hosted a "guessing game" where children recognized each other's drawings. (3/8)
We found that older children were better able to recognize all kinds of drawings. But older children were also differentially sensitive to the presence of diagnostic visual features in these drawings. (6/8)
Overall, we think these results call for systematic exploration of how different kinds of experience change visual production and recognition. Our project repository is and we’ll be publicly releasing the full dataset soon! Feedback welcome! (13/13)
I'm hiring a postdoc in the Vision Sciences Lab at Harvard, run jointly with
@talia_konkle
. If you're interested in using deep neural network models to understand human visual processing & human intuitive physical reasoning, please apply here
Older children’s recognizable drawings also contained more diagnostic features than did younger children’s (i.e. higher classifier evidence). Thus, older children’s drawings increasingly emphasized the relevant distinctions between categories in their drawings. (7/13)
Do children’s visual concepts develop early, or do children continue to learn what features distinguish categories (e.g. dogs vs. rabbits)? We asked if children improve at 1) emphasizing distinctive features when drawing & 2) using these features when recognizing drawings (2/13)
Of course, these high-resolution images are still challenging for machine-learning models—but they’re dramatically better than the lower-resolution images from prior cameras.
Here are some images with off-the-shelf Mask-R CNN segmentations superimposed. (8/11)
The BabyView captures a large, portrait-oriented vertical field-of-view that encompasses both children’s interactions with objects and with their social partners and has an effective view angle of 100° vertical by 75° horizontal. (7/11)
To do so, we collected drawings of 12 familiar objects from 4–9-year-olds (N=253) in San Jose and Beijing; each child completed 6 picture-cued drawings, 6 verbal-cued drawings, and 2 shape tracings. Then, we asked adults to guess the category depicted in each drawing. (3/9)
How do we compare the scale of language learning input for large language models vs. humans? I've been trying to come to grips with recent progress in AI. Let me explain these two illustrations I made to help. 🧵
Head-mounted cameras provide a rich and comprehensive view of what infants and young children see during their everyday experiences.
But variation between these devices in both device resolution and field-of-view has limited the field’s ability to compare results. (3/11)
These results suggest visual production and recognition of object categories improve throughout middle childhood, pointing towards refinements in children's ability to connect internal and external representations of visual concepts. What could be driving these changes? (11/13)
Further, the video data captured by all cameras to date has been relatively low-resolution.
This limits how well machine learning algorithms can operate over these rich video data. (4/11)
In this paper, we extracted the high-level visual features for each drawing using a CNN trained to recognize objects in photographs (VGG-19), and asked how well these features could be used to estimate the category that children intended to draw. (3/6)
Finally, if a new lab hopes to adopt the head-camera method for a particular developmental population, there is no single device and procedure that is widely available and easy to adopt—thus, labs must constantly “reinvent the wheel.” (5/11)
We were inspired by the work of Caitlin Fausey
@UOLearningLab
revealing changes in how infants see faces and hands across early development—and by
@JohnFranchak
’s work documenting how both infant and caregiver posture jointly shape what’s in view for infants. (3/15)
Here, we present our design process for the BabyView Camera, which collects high-resolution video, accelerometer, and gyroscope data from children approximately 6–30 months of age via a GoPro camera custom mounted on a soft child-safety helmet. (6/11)
We extracted feature representations for all drawings using a deep CNN optimized for object recognition, and then trained a classifier using these features to predict the category children were asked to draw for iterative subsets of the data. (4/8)
We found that drawings became increasingly recognizable across childhood, and neither measures of effort (e.g., time spent drawing) nor estimates of children’s tracing ability explained away this trend. (6/13)
It’s also a testament to our wonderful partnership with the Children’s Discovery Museum of San Jose (
@PurpleMuseum
), where we installed a free-standing drawing station that you can still visit today! Over two years, we collected >37,000 drawings made by 2-10 year old children.
We installed a free-standing drawing station at the Children’s Discovery Museum in San Jose (
@PurpleMuseum
), where over two years we collected >37,000 drawings from 48 categories, drawn by children 2-10 years of age. (3/13)
Children’s unrecognizable drawings still contained rich information—both animacy and object size were decodable in drawings made by children of all ages (consistent with my work with
@taliakonkle
suggesting animacy/size can be inferred without basic-level identification). (9/13)
We also present all of our protocols for interacting with families (including detailed instructions and customizable PDFs) as well as for data management and video pre-processing.
And we will continue to update these documents and scripts as we make any changes. (9/11)
Infants’ posture and orientation to their caregiver changed dramatically across this age range—many 8-month-olds spent time sitting with their caregiver’s supporting their position, while older infants were often standing and exploring the playroom on their own. (8/15)
We expect that our exact design should remain usable for years to come.
However, our modular design allows for components to be easily updated — and we commit to updating our schematics and plans if certain components change or become unavailable. (10/11)
Older children's drawings were increasingly recognizable, and children’s tracing performance didn’t explain away this trend. Further, older children’s recognizable drawings contained more diagnostic features than did younger children’s (i.e. higher classifier evidence) (5/8)
@AaronHertzmann
@hardmaru
Hi! We find that children not only get better at including diagnostic features in their drawings but that children *also* get better at using these same visual features when recognizing each other's drawings
@ashleyruba
@EvelinaLeivada
Hi 👋 I don’t think this is black and white. I'm still in my postdoc (still learning and still love the work) but I really think there’s no way I would have had the same job opportunities after my PhD vs postdoc. My computational and data science skill sets have vastly expanded.
Job posting draft
#Dyslexia
#edtech
@StanfordEd
be a part of building bridges from academia to public education working with a dynamic and collaborative team creating the next generation of open-source technology to support learning differences
Grad applications are due by Dec 6th (see ) and a postdoc ad will be posted soon. I especially encourage individuals from underrepresented backgrounds to apply!
@JhendersonIMB
Well, we've found that texture-like representations do a surprisingly good job at explaining some visual cognitive behaviors and some portion of neural responses to objects. So I don't think it's that crazy that deep CNN layers will sometimes do the same thing... (1/2)
We tried making extracting neural network activations easier for you!
Here is a Google Colab that you can run to get activations for your favorite network, based on our recently published THINGSvision library.
Pytorch:
Tensorflow:
To analyze changes in children’s drawings, we first extracted feature representations using a deep CNN optimized for object recognition. We then trained a classifier using these features to predict the category children were asked to draw. (4/13)
When reviewing, I often want to ask for a direct replication of the primary findings, especially if there isn't one built in to the paper. But what if these data aren't easy to collect (e.g., developmental / fMRI data)? It's OK to recommend replication if...
Infants are engaged in learning from others from their earliest days. How often do infants see and use social cues in naturalistic learning environments? (2/15)
We then asked children (N=1789) to recognize a subset of these drawings in four different 4AFC touch-screen games. Older children were better at recognizing all kinds of drawings—and were also differentially sensitive to diagnostic visual features during recognition. (10/13)