Pith. sign in

REVIEW 3 major objections 5 minor 97 references

Personalized Emotional Intelligence in Generative AI through Symbolic Affective Reasoning

T0 review · 3 major / 5 minor · reviewed 2026-07-14 · grok-4.5

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A hybrid AI system that mines symbolic affective rules from image–emotion data can edit photos to elicit target emotions more effectively than large multimodal models while personalizing to users without fine-tuning.

desk verdict Solid hybrid systems paper with unusually large human psychophysics; the win is the integrated EmoTree/EmoMem pipeline and metrics, not a proven causal theory of affect. read the letter →

arxiv 2607.10678 v1 pith:YAFKTXFK submitted 2026-07-12 cs.AI

Qing Lin , Mengmi Zhang This is my paper · ORCID
classification cs.AI
keywords emotionalintelligenceaffectiveimageeditingsymbolicreasoningpersonalizedgenerativeAIEmoTreehumanpsychophysicsemotionlocalizationinference-timepersonalization
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper argues that emotional intelligence in machines requires more than photorealistic generation or semantic recognition: systems must identify why images evoke affect, propose interventions that preserve scene meaning, and adapt to individual preferences. It introduces EROS, which learns a hierarchical library of compositional affective motifs (the EmoTree) from a large image–emotion dataset, localizes emotion-relevant regions, and turns those motifs into localized image edits. An expandable symbolic memory (EmoMem) stores accepted and rejected motifs so personalization happens at inference time without retraining. Across six human psychophysics experiments totaling tens of thousands of trials, the authors report that EROS more reliably elicits target positive or negative valence than existing affective editors and large multimodal pipelines, while better preserving source structure and adapting to individual motif preferences. A sympathetic reader would care because the work frames personalized affective image editing as a concrete testbed for AI that can understand and shape human emotional states in applications from media to mental health.

What carries the argument

The EmoTree: a hierarchical symbolic structure of compositional visual motifs (source concept → target valence → concept, attributes, actions, and scene context) mined from image–emotion data, retrieved to plan localized edits; paired with EmoMem, an expandable accepted/rejected motif memory that personalizes retrieval at inference time without fine-tuning.

What would settle it

On held-out images outside the training emotion dataset, run the same human pairwise and interactive editing protocols: if EROS preference, first-loop success, and cumulative success fall to or below the large multimodal baseline, or if ablating the predicted emotion regions fails to change reported valence more than ablating random regions of equal size, the central claim fails.

Watch

Extended reading notes

Core claim

EROS shows that generalizable, interpretable affective rules can be mined automatically from large-scale image–emotion data and used to steer visual content toward a desired emotional valence for a specific observer, outperforming state-of-the-art affective editing methods and large multimodal model pipelines on human preference, success rate, and structural fidelity while supporting rapid personalization through an explicit memory bank rather than model fine-tuning.

Load-bearing premise

That saliency from a simple binary emotion classifier, refined into object masks, plus motifs mined offline from captions and clustering, capture the real visual causes of human affect rather than dataset-specific correlations.

Editorial extensions

If this is right

  • Affective image editing can be treated as a measurable proxy for machine emotional intelligence spanning recognition, localization, reasoning, regulation, and personalization.
  • Symbolic motif libraries plus generative backbones can outperform pure large multimodal pipelines on emotion elicitation while keeping edits localized and source-faithful.
  • User-specific affective preferences can be stored as interpretable accepted/rejected motifs and reused across related scenes without fine-tuning.
  • Positive and negative affect may be asymmetrically organized: positive motifs cluster around a smaller semantic core while negative motifs span a wider set of cues.
  • The same motif-and-memory design could extend beyond static images to other media if emotion-relevant regions and compositional rules can be defined for those modalities.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If motif memories are truly reusable, short interactive sessions could build portable emotional profiles that transfer across apps without sharing raw images.
  • Safety guardrails that blunt negative-affect generation in proprietary models may create a systematic gap that open hybrid systems fill for research and clinical simulation—but also raise misuse risk for persuasion.
  • Same-valence enhancement and neutral-to-valenced induction, which the paper flags as less explored, are natural next tests of whether the EmoTree encodes regulation rather than only valence flipping.
  • Replacing GPT-assisted offline motif construction with fully local mining would test how much the claimed symbolic knowledge depends on proprietary language models at build time.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper introduces EROS, a hybrid symbolic–deep framework for personalized affective image editing. From EmoSet, it trains a ResNet-18 valence predictor, localizes emotion-relevant regions via Grad-CAM refined by SAM (progressive τ), and mines a hierarchical EmoTree of compositional motifs (concept, attribute, action, scene) using CLIP clustering and GPT-4o/GPT-4. Motifs become template prompts for structure-preserving Stable Diffusion inpainting (same- vs cross-valence mask inversion). An expandable EmoMem stores accepted/rejected motifs for inference-time personalization without fine-tuning. Six human psychophysics experiments (33,380 trials, 296 participants) compare EROS to LMS (GPT-based pipeline), XDream, Real retrieval, and non-interactive editors on preference, validity/efficiency, IoU, prompt alignment, and contrastive fidelity (SSIM-C/L1-C). The central claim is that EROS elicits target valence more effectively while better preserving source semantics/structure and adapts to individual motif preferences.

Significance. If the empirical superiority holds, the work is a substantial contribution to affective computing and emotionally intelligent generative AI. Strengths include a clear operationalization of emotional intelligence into five capabilities, a large multi-experiment human benchmark with attention controls and bootstrapped Welch tests, explicit symbolic memory for personalization without fine-tuning, and public code/data. The hybrid design (interpretable motifs + localized editing) addresses real limitations of global style transfer and unconstrained LMM editing. Even if causal claims about Grad-CAM and GPT-mined rules are tempered, the system-level results and evaluation protocols remain useful for the field.

major comments (3)
  1. [Sec. 4.2.2, Fig. 3B] Sec. 4.2.2 and Fig. 3B: Grad-CAM from a ResNet-18 binary valence classifier (trained on EmoSet), refined by SAM and progressive τ∈{0.5,0.3,0.1,0}, is treated as identifying the visual causes of human affect. Human IoU (0.38 vs Human–Human 0.46) shows spatial agreement but does not establish causality versus classifier-correlated features. The same- vs cross-valence mask inversion in Sec. 4.2.4 makes this interpretation load-bearing for the editing claim. A control that edits non-salient or classifier-irrelevant regions (or uses alternative localizers) is needed to support the causal reading, or the claim should be restated as predictive localization.
  2. [Sec. 4.2.3, Fig. 5A2] Sec. 4.2.3 and Sec. 4.1: EmoTree motifs are mined offline from EmoSet via CLIP clustering (δ=0.7) and GPT-4o/GPT-4 concept/motif generation from captions; many evaluation images also come from EmoSet. Prompt alignment ~90% (Fig. 5A2) and preference gains could partly reflect dataset co-occurrences and GPT language priors rather than generalizable intervention rules. The paper should report held-out or out-of-distribution scene tests (beyond the limited MSCOCO examples in Fig. S4) and/or ablate GPT-generated motifs against non-LLM alternatives to bound this circularity risk.
  3. [Sec. 2.1.1, Sec. 4.4.1] Sec. 2.1.1 and Sec. 4.4.1: LMS underperforms especially on negative valence, which the paper attributes to safety guardrails. Because LMS uses the same diffusion backbone as EROS for isolation, residual differences in prompt style, memory representation (flat concept lists vs compositional motifs), and refusal behavior remain confounds for the claim that symbolic affective reasoning is the decisive factor. A matched-prompt or guardrail-controlled comparison (or explicit reporting of LMS refusal rates) would strengthen the superiority claim over large multimodal pipelines.
minor comments (5)
  1. [Sec. 4.2] Free parameters (δ, τ schedule, γ=0.7, hard-rejection count of 3, cross-valence similarity skip 0.8, candidate sample size 50) are stated but lack sensitivity analyses; a short appendix table would help reproducibility.
  2. [Sec. 4.6.1] SSIM-C and L1-C (Sec. 4.6.1, Fig. S7) are useful but depend on human weight maps w; clarify how same-valence role reversal of w is applied when reporting aggregate fidelity across mixed trial types.
  3. [Sec. 4.2.4, Fig. 4F] Template prompts (Fig. 4F) are acknowledged as sometimes ungrammatical; quantify how often linguistic normalization fails and whether that affects human preference.
  4. [Sec. 4.5] Participant retention after attention controls varies substantially across experiments (e.g., 79→44 in Exp-EmoPrompt); report exclusion rates and any sensitivity of main results to inclusion criteria.
  5. [Sec. 3] Discussion ethics paragraph is appropriate but brief; a short note on dual-use and consent for personalized affective memory would fit the claimed mental-health applications.

Circularity Check

1 steps flagged · score 1.0 of 10

No load-bearing circular derivation; mild shared-source reuse of EmoSet for mining EmoTree and sampling evaluation images is present but does not force the human-preference or fidelity claims by construction.

  1. other [Sec. 4.1, 4.2.3 (EmoTree construction); Sec. 4.5 (source images from EmoSet test set)]
    "Our framework builds upon EmoSet [11]... Applying the above procedure to all images in EmoSet and both target valence directions yields a large repository of structured affective rules... Unless otherwise stated, all source images were drawn from the EmoSet test set."

    EmoTree motifs and the valence predictor are mined/trained on the full EmoSet distribution; evaluation images are sampled from the same dataset's test split. This is distributional reuse rather than a by-construction identity (human preference/success/IoU metrics remain independent external measurements), so it is only mild and non-load-bearing.

full rationale

The paper's central claims (superior human preference rates, cumulative success, SSIM-C/L1-C fidelity, and EmoMem personalization) rest on new forced-choice and annotation judgments collected from 296 participants across six psychophysics protocols, not on reconstruction of EmoSet training labels. EmoTree construction (CLIP clustering at δ=0.7, BLIP captions, GPT-4o concept/motif extraction) and the ResNet-18 valence predictor are data-driven from EmoSet statistics, and source images for most experiments are drawn from the EmoSet test set; this creates a mild distributional overlap that weakens the 'generalizable' claim but does not make any reported metric equal to its input by definition. Grad-CAM masks are compared to independent human region annotations (IoU 0.38 vs Human–Human 0.46); motif prompts are judged by humans for valence alignment (~90 %); edited images are judged for target-valence success and preference against LMS/XDream/Real baselines. Personalization memories are filled exclusively from live user accept/reject feedback at inference time, not from the offline dataset. No self-definitional equations, fitted-parameter-as-prediction, load-bearing self-citation uniqueness theorems, or ansatz smuggled via own prior work appear. The derivation chain is therefore empirically self-contained against external human benchmarks; the residual score of 1 reflects only the non-load-bearing train/test source overlap.

Assumptions & free parameters 6 free parameters · 5 assumptions · 4 invented entities

The central empirical claim rests on standard CV/ML components plus several design choices and invented symbolic structures. Free parameters control clustering, localization, generation stopping, and hard rejection. Domain assumptions include binary valence grouping of EmoSet’s eight emotions and the adequacy of CLIP/caption/LLM pipelines for affective concept mining. Invented entities are the EmoTree, compositional visual motifs as executable rules, and EmoMem’s accepted/rejected banks—supported inside the paper by human preference and alignment experiments but not independently established outside this work.

free parameters (6)
  • CLIP cluster cosine threshold δ = 0.7
    Controls Level-1 semantic clustering of EmoSet; authors choose δ=0.7 after inspecting fragmentation vs breadth (Fig. S3B).
  • Grad-CAM progressive thresholds τ = {0.5, 0.3, 0.1, 0}
    Editable-region schedule τ∈{0.5,0.3,0.1,0} expands masks until emotion predictor confidence or minimum extent is met; hand-chosen intervention schedule.
  • Emotion predictor acceptance threshold γ = 0.7
    Generation loop stops when P(y_t|Ĩ) exceeds γ; set to 0.7 without external calibration study.
  • Hard-rejection consecutive count = >3 consecutive rejections
    Concepts rejected in more than three consecutive interactions enter H_hard_rej and are excluded; ad hoc personalization hyperparameter.
  • Cross-valence similarity skip threshold = 0.8
    For non-neutral cross-valence edits, motifs with source–target concept similarity >0.8 are skipped to force larger semantic change; design choice affecting retrieved rules.
  • Candidate concept sample size per motif construction = 50
    50 random target-valence concepts sampled per image for GPT-4o selection during EmoTree build; computational heuristic.
assumptions (5)
  • domain assumption Eight EmoSet discrete emotions can be collapsed into binary positive vs negative valence without losing the phenomena needed for 'emotional intelligence' evaluation.
    Sec. 4.1 groups amusement/awe/contentment/excitement vs anger/disgust/fear/sadness; all modules and metrics operate on this binary.
  • domain assumption CLIP image embeddings and cosine hierarchical clustering recover semantic neighborhoods useful for transferring affective motifs across scenes.
    Level-1 EmoTree construction and inference retrieval (Sec. 4.2.3) treat CLIP similarity as the semantic prior.
  • ad hoc to paper GPT-4o/GPT-4 can extract source concepts and compose target motifs (attributes, actions, scenes) that reflect human-interpretable affective structure rather than model-specific language priors.
    Offline EmoTree construction repeatedly queries GPT-4o/GPT-4 for concept selection and motif generation (Sec. 4.2.3); affective associations are claimed to come from EmoSet statistics but motif text is LLM-generated.
  • domain assumption A ResNet-18 binary classifier trained on EmoSet plus Grad-CAM provides a usable proxy for human emotion recognition and localization.
    Modules 1–2 and stopping criteria for generation depend on this predictor (Sec. 4.2.1–4.2.2).
  • domain assumption Human forced-choice preference under instructions that show a reference image jointly measures affective effectiveness and structural preservation as claimed.
    Exp-EmoPairwise and related protocols (Sec. 4.5) operationalize success via participant selection and SSIM-C/L1-C.
invented entities (4)
  • EmoTree (hierarchical symbolic affective knowledge structure)
    purpose: Store automatically mined compositional rules linking semantic clusters, source concepts, target valence, and target motifs for controllable editing.
    Core invented knowledge base of the paper; evaluated via prompt valence alignment and editing success, not independently established outside this work.
  • Visual motif m=(c_t, a_t, r_t, s_t) as executable affective rule
    purpose: Compose concept, attributes, actions, and scene context into template prompts that steer diffusion inpainting.
    Defines the symbolic unit of intervention; human Exp-EmoPrompt supports interpretability but the entity is paper-specific.
  • EmoMem (accepted/rejected dual memory banks with hard rejection)
    purpose: Enable inference-time personalization of motif selection without fine-tuning.
    Invented personalization mechanism; Stages 2–3 of Exp-EmoRule and Exp-EmoMotif provide within-paper human evidence only.
  • SSIM-C and L1-C contrastive fidelity metrics
    purpose: Score edits that change emotion-relevant regions while preserving the rest using human weight maps w.
    New evaluation constructs for this problem setting; useful but defined for this paper’s claims.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Personalized Emotional Intelligence in Generative AI through Symbolic Affective Reasoning." pith.science (2026). https://pith.science/paper/YAFKTXFK

@misc{pith2026260710678,
  author       = {Pith},
  title        = {Pith review of: Personalized Emotional Intelligence in Generative AI through Symbolic Affective Reasoning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YAFKTXFK}},
  note         = {Machine review of arXiv:2607.10678}
}
read the original abstract

Emotional intelligence enables humans to recognize emotions, infer their causes, reason about interventions, and modify their environment to achieve desired affective states. Despite recent advances in artificial intelligence (AI), current models remain largely limited to generating realistic content or performing semantic reasoning, with little capacity for understanding, predicting, and personalizing human emotional responses. Here we introduce Emotion-augmented geneRatiOn System (EROS), a hybrid AI framework that integrates symbolic reasoning with deep learning to enable personalized emotion augmentation through visual content. Leveraging large-scale image-emotion datasets, EROS discovers generalizable affective rules, identifies emotion-relevant image regions, and predicts context-aware visual modifications that preserve scene semantics while steering emotional responses toward desired targets. To account for individual variability, EROS incorporates an expandable memory bank that supports inference-time personalization without model fine-tuning, yielding interpretable emotional profiles and rapid adaptation to new users. Across extensive human psychophysics experiments, EROS elicits target emotional responses more effectively than state-of-the-art large multimodal models while adapting to individual affective preferences. Beyond affective computing, EROS provides a foundation for AI systems that can understand, reason about, and augment human cognitive states, with potential applications in mental health, adaptive media, education, and human-computer interaction.

Figures

Figures reproduced from arXiv: 2607.10678 by the authors.

Figure 1
Figure 1. Operationalizing emotional intelligence in humans and machines. A, A concrete example illustrating emotional intelligence in humans. Given a source image and a target emotional goal (e.g., happy), humans identify emotion-relevant elements (e.g., cool lighting), reason about the causes of the current emotional response, and selectively modify visual cues (e.g., lighting, objects, and contextual details) to achieve th… view at source ↗
Figure 2
Figure 2. EROS (Emotion-augmented geneRatiOn System), our proposed emotionally intelligent and personalized generative AI framework for emotion augmentation. A, Overview of the proposed Emotion-augmented geneRatiOn System (EROS, Sec.4.2). Given an input image I and a target valence yt, EROS operationalizes emotional intelligence through five modules. Recognition: an emotion predictor P estimates the emotional valence of I and… view at source ↗
Figure 3
Figure 3. Benchmarking EROS on emotion recognition, understanding, and reasoning. A, Valence recognition performance of EROS. Confusion matrix showing the probability of predicting positive and negative valence classes. Rows correspond to ground-truth labels and columns correspond to model predictions. Color indicates prediction probability. B1, Schematic illustration of the Exp-EmoRegion experiment (Sec.4.5.3). Given an imag… view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Constructing the EmoTree from large-scale image–emotion dataset: a symbolic repository of affective rules for emotion-aware image editing. A, Schematic illustration of the EmoTree T (Sec.4.2.3). The hierarchy organizes affective knowledge into semantic clusters Ck, sou…
Figure 5
Figure 5. Figure 5: Benchmarking EROS on iterative emotion regulation. A1, Schematic illustration of the Exp-EmoPrompt experiment (Sec.4.5.4). Participants are presented with a textual description generated from an EmoTree motif and determine whether it conveys positive or negative emotio…
Figure 6
Figure 6. Figure 6: Benchmarking personalized affective profiles in EROS. A1, Schematic illustration of Stage 2 in the Exp-EmoRule experiment (Sec.4.5.5, Stage 2). Participants are presented with three sets of concepts retrieved from the personalized memories of different methods and sele…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

97 extracted references · 9 linked inside Pith

  1. [1]

    Target articles:

    J. D. Mayer, P. Salovey, and D. R. Caruso, “Target articles:” emotional intelligence: Theory, findings, and implications”,”Psychological inquiry, vol. 15, no. 3, pp. 197–215, 2004

    2004

  2. [2]

    Human abilities: Emotional intelligence,

    J. D. Mayer, R. D. Roberts, and S. G. Barsade, “Human abilities: Emotional intelligence,”Annu. Rev. Psychol., vol. 59, no. 1, pp. 507–536, 2008

    2008

  3. [3]

    Emotional intelligence as a standard intelligence.,

    J. D. Mayer, P. Salovey, D. R. Caruso, and G. Sitarenios, “Emotional intelligence as a standard intelligence.,” 2001

    2001

  4. [4]

    The positive psychology of emotional intelligence,

    P. Salovey, J. D. Mayer, D. Caruso, and S. H. Yoo, “The positive psychology of emotional intelligence,”Emotional intelligence perspectives on educational and positive psychology, pp. 185–208, 2008

    2008

  5. [5]

    Emotional pictures and sounds: A review of multimodal interactions of emotion cues in multiple domains,

    A. B. Gerdes, M. J. Wieser, and G. W. Alpers, “Emotional pictures and sounds: A review of multimodal interactions of emotion cues in multiple domains,”Frontiers in psychology, vol. 5, p. 1351, 2014

    2014

  6. [6]

    The interplay of signs and visuals: Unveiling the symbiotic relationship between semiotics and visual communication,

    A. Travere, “The interplay of signs and visuals: Unveiling the symbiotic relationship between semiotics and visual communication,”Journal of Linguistics and Communication Studies, vol. 2, no. 3, pp. 28–40, 2023

    2023

  7. [7]

    Perceptions of perceptual symbols,

    L. W. Barsalou, “Perceptions of perceptual symbols,”Behavioral and brain sciences, vol. 22, no. 4, pp. 637–660, 1999

    1999

  8. [8]

    The shaping of social perception by stimulus and knowledge cues to human animacy,

    E. S. Cross, R. Ramsey, R. Liepelt, W. Prinz, and A. F. d. C. Hamilton, “The shaping of social perception by stimulus and knowledge cues to human animacy,”Philosophical Transactions of the Royal Society B: Biological Sciences, vol. 371, no. 1686, p. 20 150 075, 2016

    2016

Show all 97 references
  1. [9]

    A new look at emotion perception: Concepts speed and shape facial emotion recognition.,

    E. C. Nook, K. A. Lindquist, and J. Zaki, “A new look at emotion perception: Concepts speed and shape facial emotion recognition.,”Emotion, vol. 15, no. 5, p. 569, 2015. 38

    2015

  2. [10]

    Perceptions of visual and multimodal symbolic mediated social touch: Role of technology modality, relationship, and task emotional salience,

    S. Yarosh, X. Wang, and Y. Yao, “Perceptions of visual and multimodal symbolic mediated social touch: Role of technology modality, relationship, and task emotional salience,”International Journal of Human-Computer Studies, vol. 159, p. 102 757, 2022

    2022

  3. [11]

    Emoset: A large-scale visual emotion dataset with rich attributes,

    J. Yang et al., “Emoset: A large-scale visual emotion dataset with rich attributes,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 20 383–20 394

    2023

  4. [12]

    How emotions are made,

    L. F. Barrett, “How emotions are made,”Providence Book Festival, May, vol. 25, p. 2019, 2019

    2019

  5. [13]

    Appraisal processes in emotion,

    P. C. Ellsworth and K. R. Scherer, “Appraisal processes in emotion,” 2003

    2003

  6. [14]

    Psychological construction in the occ model of emotion,

    G. L. Clore and A. Ortony, “Psychological construction in the occ model of emotion,”Emotion Review, vol. 5, no. 4, pp. 335–343, 2013

    2013

  7. [15]

    Context in emotion perception,

    L. F. Barrett, B. Mesquita, and M. Gendron, “Context in emotion perception,”Current directions in psychological science, vol. 20, no. 5, pp. 286–290, 2011

    2011

  8. [16]

    Inferential emotion tracking (iet) reveals the critical role of context in emotion recognition.,

    Z. Chen and D. Whitney, “Inferential emotion tracking (iet) reveals the critical role of context in emotion recognition.,”Emotion, vol. 22, no. 6, p. 1185, 2022

    2022

  9. [17]

    Object-scene semantics correlation analysis for image emotion classification,

    Z. Zhou, Z. Zhai, H. Chen, and S. Lu, “Object-scene semantics correlation analysis for image emotion classification,”Frontiers in Neuroscience, vol. 19, p. 1 657 562, 2025

    2025

  10. [18]

    Generative models improve fairness of medical classifiers under distribution shifts,

    I. Ktena et al., “Generative models improve fairness of medical classifiers under distribution shifts,”Nature Medicine, vol. 30, no. 4, pp. 1166–1173, 2024

    2024

  11. [19]

    “it happened to be the perfect thing

    S. Siddals, J. Torous, and A. Coxon, ““it happened to be the perfect thing”: Experiences of generative ai chatbots for mental health,”npj Mental Health Research, vol. 3, no. 1, p. 48, 2024

    2024

  12. [20]

    Generative ai for designing and validating easily synthesizable and structurally novel antibiotics,

    K. Swanson et al., “Generative ai for designing and validating easily synthesizable and structurally novel antibiotics,”Nature machine intelligence, vol. 6, no. 3, pp. 338–353, 2024

    2024

  13. [21]

    Integrating curricula with replays: Its effects on continual learning,

    R. J. Tee and M. Zhang, “Integrating curricula with replays: Its effects on continual learning,” inProceedings of the AAAI Symposium Series, vol. 1, 2023, pp. 109–116

    2023

  14. [22]

    Tuned compositional feature replays for efficient stream learning,

    M. B. Talbot, R. Zawar, R. Badkundri, M. Zhang, and G. Kreiman, “Tuned compositional feature replays for efficient stream learning,”IEEE Transactions on Neural Networks and Learning Systems, vol. 36, no. 2, pp. 3300–3314, 2023

    2023

  15. [23]

    Unveiling the tapestry: The interplay of generalization and forgetting in continual learning,

    Z. Shi, J. Jing, Y. Sun, J.-H. Lim, and M. Zhang, “Unveiling the tapestry: The interplay of generalization and forgetting in continual learning,”IEEE Transactions on Neural Networks and Learning Systems, 2025

    2025

  16. [24]

    Learning to learn: How to continuously teach humans and machines,

    P. Singh et al., “Learning to learn: How to continuously teach humans and machines,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 11 708–11 719

    2023

  17. [25]

    Imagenet classification with deep convolutional neural networks,

    A. Krizhevsky, I. Sutskever, and G. E. Hinton, “Imagenet classification with deep convolutional neural networks,”Advances in neural information processing systems, vol. 25, 2012. 39

    2012

  18. [26]

    Very deep convolutional networks for large-scale image recognition,

    K. Simonyan and A. Zisserman, “Very deep convolutional networks for large-scale image recognition,”arXiv preprint arXiv:1409.1556, 2014

    2014 arXiv

  19. [27]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2016, pp. 770–778

    2016

  20. [28]

    An image is worth 16x16 words: Transformers for image recognition at scale,

    A. Dosovitskiy, “An image is worth 16x16 words: Transformers for image recognition at scale,” arXiv preprint arXiv:2010.11929, 2020

    2010 arXiv

  21. [29]

    Efficient zero-shot visual search via target and context-aware transformer,

    Z. Ding et al., “Efficient zero-shot visual search via target and context-aware transformer,”arXiv preprint arXiv:2211.13470, 2022

    2022 arXiv

  22. [30]

    Object-centric learning with cyclic walks between parts and whole,

    Z. Wang, M. Z. Shou, and M. Zhang, “Object-centric learning with cyclic walks between parts and whole,”Advances in Neural Information Processing Systems, vol. 36, pp. 9388–9408, 2023

    2023

  23. [31]

    Flow snapshot neurons in action: Deep neural networks generalize to biological motion perception,

    S. Han, Z. Wang, and M. Zhang, “Flow snapshot neurons in action: Deep neural networks generalize to biological motion perception,”Advances in Neural Information Processing Systems, vol. 37, pp. 53 732–53 763, 2024

    2024

  24. [32]

    Learning to see through a baby’s eyes: Early visual diets enable robust visual intelligence in humans and machines,

    Y. Cai, B. S. Nunna, Q. Lin, and M. Zhang, “Learning to see through a baby’s eyes: Early visual diets enable robust visual intelligence in humans and machines,”arXiv preprint arXiv:2511.14440, 2025

    2025

  25. [33]

    Label-efficient online continual object detection in streaming video,

    J. Z. Wu, D. J. Zhang, W. Hsu, M. Zhang, and M. Z. Shou, “Label-efficient online continual object detection in streaming video,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 19 246–19 255

    2023

  26. [34]

    Pose prior learner: Unsupervised categorical prior learning for pose estimation,

    Z. Wang, S. Han, and M. Zhang, “Pose prior learner: Unsupervised categorical prior learning for pose estimation,”arXiv preprint arXiv:2410.03858, 2024

    2024

  27. [35]

    Gazing at rewards: Eye movements as a lens into human and ai decision-making in hybrid visual foraging,

    B. Wang et al., “Gazing at rewards: Eye movements as a lens into human and ai decision-making in hybrid visual foraging,” inProceedings of the Computer Vision and Pattern Recognition Conference, 2025, pp. 14 810–14 823

    2025

  28. [36]

    A multimodal generative ai copilot for human pathology,

    M. Y. Lu et al., “A multimodal generative ai copilot for human pathology,”Nature, vol. 634, no. 8033, pp. 466–473, 2024

    2024

  29. [37]

    Seeing sound, hearing sight: Uncovering modality bias and conflict of ai models in sound localization,

    Y. Jia et al., “Seeing sound, hearing sight: Uncovering modality bias and conflict of ai models in sound localization,”arXiv preprint arXiv:2505.11217, 2025

    2025

  30. [38]

    Adaptive visual scene understanding: Incremental scene graph generation,

    N. Khandelwal, X. Liu, and M. Zhang, “Adaptive visual scene understanding: Incremental scene graph generation,”arXiv preprint arXiv:2310.01636, 2023

    2023 arXiv

  31. [39]

    Putting visual object recognition in context,

    M. Zhang, C. Tseng, and G. Kreiman, “Putting visual object recognition in context,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2020, pp. 12 985–12 994. 40

    2020

  32. [40]

    When pigs fly: Contextual reasoning in synthetic and natural scenes,

    P. Bomatter et al., “When pigs fly: Contextual reasoning in synthetic and natural scenes,” in Proceedings of the IEEE/CVF International Conference on Computer Vision, 2021, pp. 255–264

    2021

  33. [41]

    Reason from context with self-supervised learning,

    X. Liu, A. Sikarwar, G. Kreiman, Z. Shi, and M. Zhang, “Reason from context with self-supervised learning,”arXiv preprint arXiv:2211.12817, 2022

    2022

  34. [42]

    Affectgan: Affect-based generative art driven by semantics,

    T. Galanos, A. Liapis, and G. N. Yannakakis, “Affectgan: Affect-based generative art driven by semantics,” in2021 9th International Conference on Affective Computing and Intelligent Interaction Workshops and Demos (ACIIW), IEEE, 2021, pp. 01–07

    2021

  35. [43]

    Automated colour grading using colour distribution transfer,

    F. Piti´ e, A. C. Kokaram, and R. Dahyot, “Automated colour grading using colour distribution transfer,”Computer Vision and Image Understanding, vol. 107, no. 1-2, pp. 123–137, 2007

    2007

  36. [44]

    L. A. Gatys, A. S. Ecker, and M. Bethge,A neural algorithm of artistic style, 2015. arXiv: 1508.06576 [cs.CV]

    2015 arXiv

  37. [45]

    Clipstyler: Image style transfer with a single text condition,

    G. Kwon and J. C. Ye, “Clipstyler: Image style transfer with a single text condition,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 18 062–18 071

    2022

  38. [46]

    Affective image filter: Reflecting emotions from text to images,

    S. Weng et al., “Affective image filter: Reflecting emotions from text to images,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 10 810–10 819

    2023

  39. [47]

    Make me happier: Evoking emotions through image diffusion models,

    Q. Lin, J. Zhang, Y.-S. Ong, and M. Zhang, “Make me happier: Evoking emotions through image diffusion models,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2025, pp. 16 367–16 376

    2025

  40. [48]

    Emoedit: Evoking emotions through image manipulation,

    J. Yang et al., “Emoedit: Evoking emotions through image manipulation,” inProceedings of the Computer Vision and Pattern Recognition Conference, 2025, pp. 24 690–24 699

    2025

  41. [49]

    Emokgedit: Training-free affective injection via visual cue transformation,

    J. Zhang and B. Fan, “Emokgedit: Training-free affective injection via visual cue transformation,” arXiv preprint arXiv:2601.12326, 2026

    2026

  42. [50]

    Affective image editing: Shaping emotional factors via text descriptions,

    P. Zhang et al., “Affective image editing: Shaping emotional factors via text descriptions,” International Journal of Computer Vision, vol. 134, no. 1, p. 16, 2026

    2026

  43. [51]

    Large-scale visual sentiment ontology and detectors using adjective noun pairs,

    D. Borth, R. Ji, T. Chen, T. Breuel, and S.-F. Chang, “Large-scale visual sentiment ontology and detectors using adjective noun pairs,” inProceedings of the 21st ACM international conference on Multimedia, 2013, pp. 223–232

    2013

  44. [52]

    Visual affect around the world: A large-scale multilingual visual sentiment ontology,

    B. Jou et al., “Visual affect around the world: A large-scale multilingual visual sentiment ontology,” inProceedings of the 23rd ACM international conference on Multimedia, 2015, pp. 159–168

    2015

  45. [53]

    Developing affective lexical resources.,

    A. Valitutti, C. Strapparava, and O. Stock, “Developing affective lexical resources.,”PsychNology J., vol. 2, no. 1, pp. 61–83, 2004. 41

    2004

  46. [54]

    The nonverbal affect lexicon: Theoretical perspectives from neuropsychological studies of affect perception.,

    D. Bowers, R. M. Bauer, and K. M. Heilman, “The nonverbal affect lexicon: Theoretical perspectives from neuropsychological studies of affect perception.,”Neuropsychology, vol. 7, no. 4, p. 433, 1993

    1993

  47. [55]

    Affective computing and sentiment analysis,

    E. Cambria, D. Das, S. Bandyopadhyay, and A. Feraco, “Affective computing and sentiment analysis,” inA practical guide to sentiment analysis, Springer, 2017, pp. 1–10

    2017

  48. [56]

    Robust image sentiment analysis using progressively trained and domain transferred deep networks,

    Q. You, J. Luo, H. Jin, and J. Yang, “Robust image sentiment analysis using progressively trained and domain transferred deep networks,” inProceedings of the AAAI conference on Artificial Intelligence, vol. 29, 2015

    2015

  49. [57]

    Emobank: Studying the impact of annotation perspective and representation format on dimensional emotion analysis,

    S. Buechel and U. Hahn, “Emobank: Studying the impact of annotation perspective and representation format on dimensional emotion analysis,” inProceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics: Volume 2, Short Papers, 20...

    2017

  50. [58]

    Emotions evoked by common words and phrases: Using mechanical turk to create an emotion lexicon,

    S. Mohammad and P. Turney, “Emotions evoked by common words and phrases: Using mechanical turk to create an emotion lexicon,” inProceedings of the NAACL HLT 2010 workshop on computational approaches to analysis and generation of emotion in text, 2010, pp. 26–34

    2010

  51. [59]

    Crowdsourcing a word–emotion association lexicon,

    S. M. Mohammad and P. D. Turney, “Crowdsourcing a word–emotion association lexicon,” Computational intelligence, vol. 29, no. 3, pp. 436–465, 2013

    2013

  52. [60]

    Gpt-4o system card,

    A. Hurst et al., “Gpt-4o system card,”arXiv preprint arXiv:2410.21276, 2024

    2024 arXiv

  53. [61]

    OpenAI,Introducing openai o3 and o4-mini: Our smartest and most capable models to date with full tool access,https://openai.com/index/introducing-o3-and-o4-mini/, 2025

    2025

  54. [62]

    com / index / introducing - 4o - image-generation/, 2025

    OpenAI,Introducing 4o image generation,https : / / openai . com / index / introducing - 4o - image-generation/, 2025

    2025

  55. [63]

    Fortin, Alisa and Vernade, Guillaume and Kampf, Kat and Reshi, Ammaar,Introducing gemini 2.5 flash image, our state-of-the-art image model,https://developers.googleblog.com/en/ introducing-gemini-2-5-flash-image/, 2025

    2025

  56. [64]

    Capacity of generative ai to interpret human emotions from visual and textual data: Pilot evaluation study,

    Z. Elyoseph et al., “Capacity of generative ai to interpret human emotions from visual and textual data: Pilot evaluation study,”JMIR Mental Health, vol. 11, e54369, 2024

    2024

  57. [65]

    Do llms” feel

    C. Wang et al., “Do llms” feel”? emotion circuits discovery and control,”arXiv preprint arXiv:2510.11328, 2025

    2025

  58. [66]

    Emotion concepts and their function in a large language model,

    N. Sofroniew et al., “Emotion concepts and their function in a large language model,” Anthropic/Transformer Circuits. transformer-circuits. pub/2026/emotions, 2026

    2026

  59. [67]

    Chain-of-thought is not explainability,

    F. Barez et al., “Chain-of-thought is not explainability,”Preprint, alphaXiv, p. v1, 2025

    2025

  60. [68]

    Gpt-4v with emotion: A zero-shot benchmark for generalized emotion recognition,

    Z. Lian et al., “Gpt-4v with emotion: A zero-shot benchmark for generalized emotion recognition,”Information Fusion, vol. 108, p. 102 367, 2024. 42

    2024

  61. [69]

    Research can help to tackle ai-generated disinformation,

    S. Feuerriegel et al., “Research can help to tackle ai-generated disinformation,”Nature Human Behaviour, vol. 7, no. 11, pp. 1818–1821, 2023

    2023

  62. [70]

    Emotionhallucer: Evaluating emotion hallucinations in multimodal large language models,

    B. Xing et al., “Emotionhallucer: Evaluating emotion hallucinations in multimodal large language models,”arXiv preprint arXiv:2505.11405, 2025

    2025 arXiv

  63. [71]

    Signs of consciousness in ai: Can gpt-3 tell how smart it really is?

    L. Boji´ c, I. Stojkovi´ c, and Z. Joli´ c Marjanovi´ c, “Signs of consciousness in ai: Can gpt-3 tell how smart it really is?”Humanities and Social Sciences Communications, vol. 11, no. 1, pp. 1–15, 2024

    2024

  64. [72]

    Language models don’t always say what they think: Unfaithful explanations in chain-of-thought prompting,

    M. Turpin, J. Michael, E. Perez, and S. Bowman, “Language models don’t always say what they think: Unfaithful explanations in chain-of-thought prompting,”Advances in Neural Information Processing Systems, vol. 36, pp. 74 952–74 965, 2023

    2023

  65. [73]

    A survey on hallucination in large language models: Principles, taxonomy, challenges, and open questions,

    L. Huang et al., “A survey on hallucination in large language models: Principles, taxonomy, challenges, and open questions,”ACM Transactions on Information Systems, vol. 43, no. 2, pp. 1–55, 2025

    2025

  66. [74]

    The emotional intelligence of the gpt-4 large language model,

    G. D. Vzorinab, A. M. Bukinichac, A. V. Sedykha, I. I. Vetrovab, and E. A. Sergienkob, “The emotional intelligence of the gpt-4 large language model,”Psychology in Russia: State of the art, vol. 17, no. 2, pp. 85–99, 2024

    2024

  67. [75]

    Hallucination of multimodal large language models: A survey,

    Z. Bai et al., “Hallucination of multimodal large language models: A survey,”arXiv preprint arXiv:2404.18930, 2024

    2024 arXiv

  68. [76]

    The other you in black mirror: First steps from chatbots to personalized llm clones,

    M. Sun, M. Zhang, and G. Kreiman, “The other you in black mirror: First steps from chatbots to personalized llm clones,” 2025

    2025

  69. [77]

    Ai-based personalized e-learning systems: Issues, challenges, and solutions,

    M. Murtaza, Y. Ahmed, J. A. Shamsi, F. Sherwani, and M. Usman, “Ai-based personalized e-learning systems: Issues, challenges, and solutions,”IEEE access, vol. 10, pp. 81 323–81 342, 2022

    2022

  70. [78]

    Exploring the impact of artificial intelligence application in personalized learning environments: Thematic analysis of undergraduates’ perceptions in china,

    X. Wang, X. Xu, Y. Zhang, S. Hao, and W. Jie, “Exploring the impact of artificial intelligence application in personalized learning environments: Thematic analysis of undergraduates’ perceptions in china,”Humanities and Social Sciences Communications, vol. 11, no. 1, pp. 1–10, 2024

    2024

  71. [79]

    Generative ai and gamification for personalized learning: Literature review and future challenges,

    F. Abbes, S. Bennani, and A. Maalel, “Generative ai and gamification for personalized learning: Literature review and future challenges,”SN Computer Science, vol. 5, no. 8, p. 1154, 2024

    2024

  72. [80]

    Ai-generated characters for supporting personalized learning and well-being,

    P. Pataranutaporn et al., “Ai-generated characters for supporting personalized learning and well-being,”Nature Machine Intelligence, vol. 3, no. 12, pp. 1013–1022, 2021

    2021

  73. [81]

    Closing the accessibility gap to mental health treatment with a personalized self-referral chatbot,

    J. Habicht et al., “Closing the accessibility gap to mental health treatment with a personalized self-referral chatbot,”Nature medicine, vol. 30, no. 2, pp. 595–602, 2024

    2024

  74. [82]

    Daga, Soham and Sreedhar, Sreeram and Shah, Dhravya,Supermemory is the new state-of-the-art in agent memory,https://supermemory.ai/research/. 43

  75. [83]

    Instructpix2pix: Learning to follow image editing instructions,

    T. Brooks, A. Holynski, and A. A. Efros, “Instructpix2pix: Learning to follow image editing instructions,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 18 392–18 402

    2023

  76. [84]

    Sdedit: Guided image synthesis and editing with stochastic differential equations,

    C. Meng et al., “Sdedit: Guided image synthesis and editing with stochastic differential equations,”arXiv preprint arXiv:2108.01073, 2021

    2021 arXiv

  77. [85]

    Adding conditional control to text-to-image diffusion models,

    L. Zhang, A. Rao, and M. Agrawala, “Adding conditional control to text-to-image diffusion models,” inProceedings of the IEEE/CVF international conference on computer vision, 2023, pp. 3836–3847

    2023

  78. [86]

    Blip-diffusion: Pre-trained subject representation for controllable text-to-image generation and editing,

    D. Li, J. Li, and S. Hoi, “Blip-diffusion: Pre-trained subject representation for controllable text-to-image generation and editing,”Advances in Neural Information Processing Systems, vol. 36, pp. 30 146–30 166, 2023

    2023

  79. [87]

    Tolstoy,Anna karenina

    L. Tolstoy,Anna karenina. Lulu. com, 2016

    2016

  80. [88]

    Grad-cam: Visual explanations from deep networks via gradient-based localization,

    R. R. Selvaraju et al., “Grad-cam: Visual explanations from deep networks via gradient-based localization,” inProceedings of the IEEE international conference on computer vision, 2017, pp. 618–626

    2017

  81. [89]

    Segment anything,

    A. Kirillov et al., “Segment anything,” inProceedings of the IEEE/CVF international conference on computer vision, 2023, pp. 4015–4026

    2023

  82. [90]

    Blip: Bootstrapping language-image pre-training for unified vision-language understanding and generation,

    J. Li, D. Li, C. Xiong, and S. Hoi, “Blip: Bootstrapping language-image pre-training for unified vision-language understanding and generation,” inInternational conference on machine learning, PMLR, 2022, pp. 12 888–12 900

    2022

  83. [91]

    Spacy: Industrial-strength natural language processing in python,

    M. Honnibal, I. Montani, S. Van Landeghem, A. Boyd, et al., “Spacy: Industrial-strength natural language processing in python,” 2020

    2020

  84. [92]

    High-resolution image synthesis with latent diffusion models,

    R. Rombach, A. Blattmann, D. Lorenz, P. Esser, and B. Ommer, “High-resolution image synthesis with latent diffusion models,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 10 684–10 695

    2022

  85. [93]

    Evolving images for visual neurons using a deep generative network reveals coding principles and neuronal preferences,

    C. R. Ponce et al., “Evolving images for visual neurons using a deep generative network reveals coding principles and neuronal preferences,”Cell, vol. 177, no. 4, pp. 999–1009, 2019

    2019

  86. [94]

    Learning transferable visual models from natural language supervision,

    A. Radford et al., “Learning transferable visual models from natural language supervision,” in International conference on machine learning, PMLR, 2021, pp. 8748–8763

    2021

  87. [95]

    Amazon mechanical turk,

    A. M. Turk, “Amazon mechanical turk,”Retrieved August, vol. 17, p. 2012, 2012

    2012

  88. [96]

    Image quality assessment: From error visibility to structural similarity,

    Z. Wang, A. C. Bovik, H. R. Sheikh, and E. P. Simoncelli, “Image quality assessment: From error visibility to structural similarity,”IEEE transactions on image processing, vol. 13, no. 4, pp. 600–612, 2004

    2004

  89. [97]

    Microsoft coco: Common objects in context,

    T.-Y. Lin et al., “Microsoft coco: Common objects in context,” inEuropean conference on computer vision, Springer, 2014, pp. 740–755. 44 Main Figures Figure 1:Operationalizing emotional intelligence in humans and machines. A, A concrete example illustrating emotional intellige...

    2014

Pith tools

Reviewed July 14, 2026 · model on record in the stance chip above.