All filters off — toggle a chip or lower the importance slider to see nodes.
Top hubs · by degree
Legend
concept
claim
result
method
entity
MAP
Interactive version —
how to use this graph
✓
fast mental map
Click ▶ Guided tour for a 60-second walk through the editor's pick. Or hover any node to focus; click for source; ★ nodes you want to come back to; ⌘+click two nodes to compare.
✓
share a specific view
Select any node, copy URL — the link encodes selection, zoom, and filters. Save it as a named view (⌘ views). Annotations save locally per paper. </> embed generates an iframe.
✗
not a citable source
Do not quote the graph as an authority. Edge labels and importance scores are interpretive judgments by the generating agent. Any claim worth citing must be traced back to the original paper.
reliability noteHeadline structure and importance-5 nodes are stable across runs. Mid-tier nodes (importance 2–3) and edge type distinctions are interpretive and may differ between runs. Click any node to see its source citation — nodes marked "training memory" or "inferred" were not directly verified against the source document.
LOOMUS™ and the Knowledge-Loom methodology are proprietary. Visual system is original to LOOMUS.
Knowledge Graph: The Worlds I See: Curiosity, Exploration, and Discovery at the Dawn of AI (Fei-Fei Li, 2023)
Editorial spotlight: ImageNet → the paradigm shift that made modern AI possible
Concepts
Li's human-centered AI vision (importance 5): The book's central argument: AI must serve human values, not replace human judgment. Technology shaped by ethics and care.. Source: (from training memory of book).
WordNet as ImageNet blueprint (importance 4): George Miller's linguistic ontology inspires the idea of a visual ontology at scale.. Source: (from training memory of book).
Li's 'ambient intelligence' for eldercare (importance 4): Vision-based sensors in homes to help aging populations live independently. Privacy-preserving, dignity-centered design.. Source: (from training memory of book).
Immigrant 'outsider' lens on AI (importance 4): Fei-Fei credits immigrant experience with giving her non-traditional perspective: questioning norms, seeing gaps others miss.. Source: (from training memory of book).
representation learning via labels (importance 4): Core insight: networks learn generalizable features when trained on diverse labeled data at scale.. Source: (from training memory of book).
AI for healthcare (Li's later focus) (importance 4): Vision systems for hospital patient monitoring, eldercare, early disease detection. Human-centered applications.. Source: (from training memory of book).
Li's 'responsible AI' framework (importance 4): AI development must include ethicists, social scientists, affected communities from design stage onward.. Source: (from training memory of book).
interdisciplinary AI research model (importance 4): HAI's core principle: CS alone cannot build beneficial AI. Must integrate humanities, medicine, law, ethics.. Source: (from training memory of book).
ImageNet as open data precedent (importance 4): Dataset released freely. Democratizes AI research, enables global participation. Contrast with proprietary industry data.. Source: (from training memory of book).
human visual intelligence as inspiration (importance 4): Fei-Fei's motivation: humans learn to see effortlessly from sparse data. Can AI do the same with right dataset?. Source: (from training memory of book).
'AI for good' movement (Li as leader) (importance 4): Post-2015 push for AI applications in social good: healthcare, climate, education. Li's public advocacy role.. Source: (from training memory of book).
dignity as design principle (importance 4): Eldercare tech must preserve autonomy, not infantilize. Fei-Fei's insistence on user agency in assistive systems.. Source: (from training memory of book).
cautious optimism about AI future (importance 4): Book ends on guarded hope: if we center human values, AI can be net good. But requires vigilance, course correction.. Source: (from training memory of book).
AI winters (1970s, 1990s) (importance 3): Historical periods of AI hype collapse. Fei-Fei enters field during skepticism about neural networks' viability.. Source: (from training memory of book).
object recognition as AI grand challenge (importance 3): Pre-ImageNet consensus: recognizing objects in natural images was decades away. ImageNet collapses that timeline.. Source: (from training memory of book).
from objects to scene understanding (importance 3): Post-ImageNet research direction: not just 'what' but 'where', 'how', relationships between entities.. Source: (from training memory of book).
gender barriers in AI/CS (importance 3): Fei-Fei as one of few prominent women in AI. Advocates for inclusive culture, mentors women students.. Source: (from training memory of book).
crowdwork ethics (Mechanical Turk) (importance 3): Fei-Fei grapples with low wages, repetitive labor of MTurk workers. Tension between scale goals and fair compensation.. Source: (from training memory of book).
privacy in ambient vision systems (importance 3): Eldercare sensors must not surveil. Design principle: extract activity signals without storing identifiable images.. Source: (from training memory of book).
need for explainable AI (healthcare) (importance 3): Medical applications require interpretability. Black-box models insufficient when human lives at stake.. Source: (from training memory of book).
awareness of 'data colonialism' (importance 3): Western-centric datasets encode cultural biases. Li advocates for globally representative data collection.. Source: (from training memory of book).
managing AI hype vs. reality (importance 3): Post-2012 media frenzy around AI. Fei-Fei tries to balance public excitement with realistic expectations.. Source: (from training memory of book).
navigating Chinese-American identity (importance 3): Fei-Fei's dual cultural fluency. Asset in global AI collaboration, but also source of 'othering' in US academia.. Source: (from training memory of book).
Li's mentorship philosophy (importance 3): Students as whole people, not just research output. Advocates work-life balance, mental health support in academia.. Source: (from training memory of book).
calls for AI regulation (policy) (importance 3): Li testifies to Congress, advises policymakers. Argues for proactive regulation before harms scale.. Source: (from training memory of book).
US-China AI competition concerns (importance 3): Fei-Fei warns against zero-sum framing. Advocates for international collaboration on AI safety, ethics.. Source: (from training memory of book).
curiosity as research driver (importance 3): Book's autobiographical thread: following intellectual curiosity over careerism. Non-linear path as strength.. Source: (from training memory of book).
mother-daughter relationship (care ethics) (importance 3): Fei-Fei's relationship with aging mother shapes eldercare AI vision. Personal experience grounds technical work.. Source: (from training memory of book).
parents' sacrifice narrative (importance 3): Parents give up professional careers in China for children's future. Motivates Fei-Fei's drive to 'make it worth it'.. Source: (from training memory of book).
vision for embodied AI (robotics) (importance 3): Post-ImageNet research direction: vision for robots interacting with physical world. Builds on recognition foundation.. Source: (from training memory of book).
motherhood + tenure-track struggle (importance 3): Fei-Fei navigates childcare, research demands, cultural expectations. Advocates for structural support for parent-researchers.. Source: (from training memory of book).
benchmark-driven research culture (importance 3): ImageNet competition creates 'leaderboard' mentality. Pro: measurable progress. Con: narrow optimization, gaming metrics.. Source: (from training memory of book).
vision + language integration (future direction) (importance 3): Post-ImageNet frontier: models that understand images AND text together. Precursor to modern multimodal AI.. Source: (from training memory of book).
AI for disability/accessibility (importance 3): Vision systems for blind users, voice control for mobility-impaired. Li advocates research investment here.. Source: (from training memory of book).
resistance to surveillance capitalism (importance 3): Fei-Fei critiques business models built on invasive data collection. Calls for regulation, ethical alternatives.. Source: (from training memory of book).
AI literacy education imperative (importance 3): Public must understand AI capabilities/limits to demand accountability. Li supports K-12, public education initiatives.. Source: (from training memory of book).
AI development in Global South (importance 3): Li warns against AI gap widening inequality. Advocates for capacity-building, data sovereignty in developing countries.. Source: (from training memory of book).
neuroscience as AI inspiration (Koch, Hubel/Wiesel) (importance 3): Visual cortex research inspires hierarchical feature learning in CNNs. Fei-Fei's early neuroscience exposure.. Source: (from training memory of book).
science communication as storytelling (importance 3): Book itself models narrative approach to technical work. Fei-Fei believes stories make science accessible, human.. Source: (from training memory of book).
responsibility to next generation of AI researchers (importance 3): Li's mentorship focus: training students who will build AI differently, with ethics embedded from the start.. Source: (from training memory of book).
AI research reproducibility issues (importance 2): Proprietary datasets, undisclosed hyperparameters hinder verification. ImageNet's openness as counter-model.. Source: (from training memory of book).
imposter syndrome (immigrant, woman in STEM) (importance 2): Fei-Fei's recurring self-doubt despite accomplishments. Shares to normalize experience, advocate for support systems.. Source: (from training memory of book).
Claims
Li's 'data hunger' thesis (2007) (importance 5): Recognition that AI progress was bottlenecked by lack of large-scale labeled data, not algorithms.. Source: (from training memory of book).
ImageNet enables deep learning revival (importance 5): The dataset's scale allowed neural networks to shine; pre-ImageNet, models were too small to demonstrate their power.. Source: (from training memory of book).
Li's 'North Star' = human flourishing (importance 5): Core thesis: technology choices should optimize for human dignity, agency, well-being. Not efficiency or profit alone.. Source: (from training memory of book).
data-centric (not model-centric) breakthrough (importance 5): ImageNet's lesson: better data > better algorithms (at least until you have enough data). Shifts field's focus.. Source: (from training memory of book).
recognition of dataset bias problem (importance 4): ImageNet reflects biases of internet image distribution, labelers' cultural assumptions. Li advocates for conscious bias mitigation.. Source: (from training memory of book).
2016-2018 AI ethics awakening (importance 4): Field-wide reckoning with facial recognition abuse, military applications, bias. Fei-Fei's voice in demanding accountability.. Source: (from training memory of book).
2012 as AI's inflection point (importance 4): AlexNet's ImageNet win marks before/after. Pre-2012: AI winter skepticism. Post-2012: deep learning everywhere.. Source: (from training memory of book).
empirical discovery of data scaling laws (importance 4): ImageNet demonstrates: more data → better performance, in ways smaller datasets couldn't reveal. Foreshadows LLM era.. Source: (from training memory of book).
ImageNet's dual legacy (power + problems) (importance 4): Enabled AI revolution, but also perpetuated biases, raised privacy concerns. Li grapples with unintended consequences.. Source: (from training memory of book).
Li's legacy anxiety (ImageNet's double-edged impact) (importance 4): Final chapters grapple with: will ImageNet be remembered for enabling progress or enabling harms? Still unresolved.. Source: (from training memory of book).
ImageNet's problematic categories (importance 3): Dataset includes offensive person-labeling categories. 2019-2020 critiques lead to category removal, dataset revision.. Source: (from training memory of book).
academia-industry AI talent drain (importance 3): Post-ImageNet, tech companies hire away top researchers. Threatens university research capacity.. Source: (from training memory of book).
physics training shapes AI approach (importance 3): Fei-Fei credits physics background with systems thinking, mathematical rigor, questioning first principles.. Source: (from training memory of book).
fear of new AI winter (if hype exceeds reality) (importance 3): Li warns: over-promising AI capabilities → public backlash → funding cuts. Need honest communication about limitations.. Source: (from training memory of book).
environmental cost of AI training (importance 2): Large-scale model training emits significant CO2. Li acknowledges need for sustainable AI research practices.. Source: (from training memory of book).
gender/race bias in peer review (importance 2): Fei-Fei experiences dismissal of ideas, later accepted when proposed by male colleagues. Systemic problem.. Source: (from training memory of book).
lack of role models (Asian women in AI) (importance 2): Fei-Fei's generation had almost no visible Asian women AI leaders. She consciously steps into that role for next generation.. Source: (from training memory of book).
Empirical results
ImageNet (2009): 14M images, 22K categories (importance 5): The massive labeled dataset that enabled the deep learning revolution. Fei-Fei's core contribution to AI history.. Source: (from training memory of book).
AlexNet ImageNet victory (2012) (importance 5): Hinton's team wins ImageNet competition with deep learning, validating the dataset's transformative role.. Source: (from training memory of book).
Visual Genome dataset (importance 3): Follow-up to ImageNet: images annotated with relationships, not just labels. Moves toward scene understanding.. Source: (from training memory of book).
Methods
Amazon Mechanical Turk for labeling (importance 4): Crowdsourcing strategy to label millions of images cheaply. Controversial but necessary for scale.. Source: (from training memory of book).
ImageNet Large Scale Visual Recognition Challenge (ILSVRC) (importance 4): Annual competition (2010-2017) that became the benchmark for computer vision progress.. Source: (from training memory of book).
ImageNet pre-training → transfer learning (importance 4): Networks trained on ImageNet transfer to other vision tasks (medical imaging, robotics). Unlocks new research paradigms.. Source: (from training memory of book).
convolutional neural networks (CNNs) (importance 3): Architecture that AlexNet used to win ImageNet. Not invented by Li but enabled by her dataset.. Source: (from training memory of book).
GPU acceleration for training (importance 3): AlexNet's use of GPUs makes training on ImageNet-scale data feasible. Hardware + data convergence.. Source: (from training memory of book).
ImageNet annotation pipeline design (importance 3): Multi-stage verification: initial labeling, quality checks, consensus voting. Balances speed with accuracy.. Source: (from training memory of book).
ImageNet ontology (WordNet hierarchy) (importance 3): Hierarchical category structure borrowed from WordNet. Enables coarse-to-fine recognition, semantic relationships.. Source: (from training memory of book).
ICU patient monitoring (vision AI) (importance 3): Cameras track patient movement, detect falls, alert staff. Privacy-preserving design: no face identification.. Source: (from training memory of book).
visual attention models (pre-ImageNet work) (importance 2): Fei-Fei's early PhD research. Modeling where humans look in images. Informs later scene understanding work.. Source: (from training memory of book).
Bayesian models of recognition (Caltech era) (importance 2): Pre-deep-learning approach. Li's early work on hierarchical probabilistic vision. Superseded by neural nets at scale.. Source: (from training memory of book).
Entities
Chengdu childhood (1976-1992) (importance 4): Fei-Fei's formative years in China during cultural shifts, before emigrating to US.. Source: (from training memory of book).
Stanford HAI co-founding (2019) (importance 4): Human-Centered AI Institute. Institutionalizes the vision that AI must be designed with human welfare at center.. Source: (from training memory of book).
Parsippany, NJ: immigrant struggle (importance 3): Family arrives in US. Parents work menial jobs; Fei-Fei navigates American school system while working at dry cleaner.. Source: (from training memory of book).
Princeton physics → AI pivot (importance 3): Undergraduate physics major, discovers AI through Christof Koch's visual neuroscience work.. Source: (from training memory of book).
Caltech PhD (Pietro Perona) (importance 3): Trains in computer vision under Perona. Early work on Bayesian models of visual recognition.. Source: (from training memory of book).
Stanford AI Lab directorship (2013-2018) (importance 3): Fei-Fei becomes first woman to direct Stanford AI Lab. Builds research culture around vision + healthcare.. Source: (from training memory of book).
Google Cloud AI/ML Chief Scientist (2017-2018) (importance 3): Brief industry tenure. Advocates for AI ethics within Google during tensions around military contracts.. Source: (from training memory of book).
Google Project Maven controversy (importance 3): Pentagon drone imagery contract. Fei-Fei internally raises concerns; thousands of Google employees protest.. Source: (from training memory of book).
Mother's illness → eldercare motivation (importance 3): Personal experience with parent's health decline motivates ambient intelligence research direction.. Source: (from training memory of book).
Christof Koch (visual neuroscience mentor) (importance 3): Caltech neuroscientist. His work on visual attention draws Fei-Fei from physics into vision science.. Source: (from training memory of book).
Pietro Perona (PhD advisor) (importance 3): Caltech computer vision pioneer. Guides Fei-Fei's early probabilistic vision work. Later collaborates on Caltech-101.. Source: (from training memory of book).
George Miller (WordNet creator) (importance 3): Princeton cognitive psychologist. WordNet's hierarchical structure becomes the model for ImageNet's ontology.. Source: (from training memory of book).
Geoffrey Hinton (AlexNet supervisor) (importance 3): Deep learning pioneer. His student Alex Krizhevsky wins 2012 ImageNet, validating neural networks at scale.. Source: (from training memory of book).
facial recognition moratorium advocacy (importance 3): Li supports bans on police use of face recognition. Technology's accuracy insufficient, bias too severe.. Source: (from training memory of book).
PASCAL VOC (pre-ImageNet benchmark) (importance 2): Earlier computer vision dataset. ~20K images, 20 categories. Too small for deep learning to work.. Source: (from training memory of book).
Caltech-101 dataset (importance 2): Perona's lab's earlier dataset. 101 object categories. Fei-Fei saw the need to scale orders of magnitude beyond this.. Source: (from training memory of book).
High school dry cleaner work (importance 2): After-school job helping parents. Fei-Fei does homework between customers. Formative class-consciousness.. Source: (from training memory of book).
Cultural Revolution aftermath (China) (importance 2): Parents' generation scarred by Mao era. Shapes family's decision to emigrate and Fei-Fei's awareness of ideology's dangers.. Source: (from training memory of book).
Silvio Savarese (husband, collaborator) (importance 2): Computer vision researcher. Personal and professional partner through ImageNet era.. Source: (from training memory of book).
ImageNet Roulette art project (2019) (importance 2): Artists expose ImageNet's offensive person categories via public demo. Catalyzes dataset reform.. Source: (from training memory of book).
COCO dataset (Microsoft) (importance 2): Competitor/successor to ImageNet. Context understanding, not just object labels. Reflects evolution of vision research.. Source: (from training memory of book).
Li's TED talk (2015): 'How we teach computers to understand pictures' (importance 2): Public-facing explanation of ImageNet's significance. Millions of views; raises Fei-Fei's public profile.. Source: (from training memory of book).
pre-tenure years (Illinois, Princeton, Stanford) (importance 2): Fei-Fei builds ImageNet during assistant professor grind. Risky bet on unfunded, unglamorous data work.. Source: (from training memory of book).
NSF grant rejections for ImageNet (importance 2): Early funding proposals fail. Reviewers don't see value in 'just' building a dataset. Li persists anyway.. Source: (from training memory of book).
Stanford AI Lab culture shift (importance 2): Fei-Fei's directorship emphasizes collaboration, diversity, ethics alongside technical excellence. Contrast with earlier 'hacker' culture.. Source: (from training memory of book).
ImageNet paper acceptance struggles (importance 2): Early ImageNet papers face skeptical reviewers. 'Just a dataset' critique. Eventually gains recognition as foundational.. Source: (from training memory of book).
CVPR/ICCV computer vision community (importance 2): The conference ecosystem where ImageNet gains traction. Fei-Fei's network building in this community.. Source: (from training memory of book).
ImageNet team (grad students, postdocs) (importance 2): Jia Deng, Olga Russakovsky, others. Collaborative effort often erased in 'genius inventor' narratives. Li credits team.. Source: (from training memory of book).
AI for cancer detection (Stanford collaboration) (importance 2): Transfer learning from ImageNet to medical imaging. Early success story for AI-for-good applications.. Source: (from training memory of book).