← Roland Memisevic

Publications

Full chronological list. For an authoritative, continuously-updated view with citation counts, see Google Scholar →.

2026

  1. Multiplicative vs. additive state coupling in learned optimizer schedules
    Langston, M. H., Letourneau, P. D., Jones, D., Memisevic, R., Scott, P., Lee, M., …
    NeurIPS 2026 OPT-ML Workshop
  2. Graph coarsening enables escaping local optima in compute-constrained nonconvex MAX-CUT
    Langston, M. H., Letourneau, P. D., Jones, D., Teague, H., Scott, P., Lee, M., …, Memisevic, R.
    NeurIPS 2026 OPT Workshop
  3. Alleviating hallucination in reasoning tasks with training-free uncertainty-guided steering
    Liu, L., Hou, Q., Jian, Y., Pourreza, R., Ghavamzadeh, M., Memisevic, R., Qin, Y., Cai, H.
    Neural Information Processing Systems (NeurIPS 2026)
  4. On locality and length generalization in visual reasoning
    Madan, P., Haresh, S., Ebrahimi, R., Panchal, S., Bhattacharyya, A., …, Memisevic, R.
    European Conference on Computer Vision (ECCV 2026)
  5. Streaming interventions: Can video large language models correct mistakes as they occur?
    Bhattacharyya, A., Mahajan, S., Haresh, S., Yasarla, R., Pourreza, R., Liu, L., …, Memisevic, R.
    Neural Information Processing Systems (NeurIPS 2026)
  6. Notes-to-Self: Scratchpad augmented VLAs for memory-dependent manipulation tasks
    Haresh, S., Dijkman, D., Bhattacharyya, A., Memisevic, R.
    International Conference on Robotics and Automation (ICRA 2026)
  7. On the “induction bias” in sequence models
    Ebrahimi, M. R., Defferrard, M., Panchal, S., Memisevic, R.
    International Conference on Machine Learning (ICML 2026)
  8. From out-of-distribution detection to hallucination detection: A geometric view
    Liu, L., Pourreza, R., Jian, Y., Qin, Y., Memisevic, R.
    International Conference on Machine Learning (ICML 2026)
  9. Can vision-language models answer face-to-face questions in the real world?
    Pourreza, R., Dagli, R., Bhattacharyya, A., Panchal, S., Berger, G., Memisevic, R.
    International Conference on Learning Representations (ICLR 2026)
  10. Enhancing hallucination detection through noise injection
    Liu, L., Pourreza, R., Panchal, S., Bhattacharyya, A., Jian, Y., Qin, Y., …, Memisevic, R.
    International Conference on Learning Representations (ICLR 2026)

2025

  1. Can multi-modal LLMs provide live step-by-step task guidance?
    Bhattacharyya, A., Xu, B., …, Sigal, L., Memisevic, R.
    Neural Information Processing Systems (NeurIPS 2025)
  2. Revisiting bi-linear state transitions in recurrent neural networks
    Ebrahimi, M. R., Memisevic, R.
    Neural Information Processing Systems (NeurIPS 2025)
  3. Multi-draft speculative sampling: Canonical decomposition and theoretical limits
    Khisti, A., Ebrahimi, M. R., Dbouk, H., Behboodi, A., Memisevic, R., Louizos, C.
    International Conference on Learning Representations (ICLR 2025)
  4. Delayed attention training improves length generalization in transformer–RNN hybrids
    Phan, B., Ebrahimi, M. R., Haresh, S., Memisevic, R.
    NeurIPS 2025 Workshop on What Transformers Can(not) Do
  5. Replacing thinking with tool usage enables reasoning in small language models
    Rainone, C., Bakker, T., Memisevic, R.
    arXiv:2507.05065
  6. Grounding code understanding in step-by-step execution
    Zhang, D. W., Defferrard, M., Rainone, C., Memisevic, R.
    International Conference on Learning Representations (ICLR 2025)

2024

  1. ClevrSkills: Compositional language and visual reasoning in robotics
    Haresh, S., Dijkman, D., Bhattacharyya, A., Memisevic, R.
    Neural Information Processing Systems (NeurIPS 2024)
  2. What to say and when to say it: Live fitness coaching as a testbed for situated interaction
    Panchal, S., Bhattacharyya, A., Berger, G., …, Memisevic, R.
    Neural Information Processing Systems (NeurIPS 2024), Datasets & Benchmarks Track
  3. Your context is not an array: Unveiling random access limitations in transformers
    Ebrahimi, M. R., Panchal, S., Memisevic, R.
    Conference on Language Modeling (COLM 2024)
  4. AirLetters: An open video dataset of characters drawn in the air
    Dagli, R., Berger, G., Materzynska, J., Bax, I., Memisevic, R.
    European Conference on Computer Vision (ECCV 2024), HANDS Workshop
  5. Look, remember and reason: Grounded reasoning in videos with language models
    Bhattacharyya, A., Panchal, S., Pourreza, R., Lee, M., Madan, P., Memisevic, R.
    International Conference on Learning Representations (ICLR 2024)
  6. Towards bridging classical and neural computation through a read-eval-print loop
    Zhang, D. W., Defferrard, M., Rainone, C., Memisevic, R.
    ICML 2024 Workshop on LLMs and Cognition
  7. PlotTwist: Vision-language models struggle with reasoning over mathematical plots
    Madan, P., Haresh, S., Bhattacharyya, A., Liu, L., Pourreza, R., Panchal, S., Memisevic, R.
    NeurIPS 2024 EVAL-FoMo Workshop (Evaluating Foundation Models)
  8. Math-Search: A benchmark for multi-hop visual reasoning over plots
    Madan, P., Haresh, S., Bhattacharyya, A., Liu, L., Pourreza, R., Panchal, S., Memisevic, R.
    CVPR 2024 MAR Workshop (Multimodal Algorithmic Reasoning)  ·  earlier version of the PlotTwist work

2023

  1. Situated real-time interaction with a virtually embodied avatar
    Panchal, S., Berger, G., Mercier, A., Böhm, C., Dietrichkeit, F., Li, X., Pourreza, R., Madan, P., Todorovich, M., Memisevic, R.
    CVPR 2023 Embodied AI Workshop
  2. Deductive verification of chain-of-thought reasoning
    Ling, Z., Fang, Y., Li, X., Huang, Z., Lee, M., Memisevic, R., Su, H.
    Neural Information Processing Systems (NeurIPS 2023)
  3. Painter: Teaching auto-regressive language models to draw sketches
    Pourreza, R., Bhattacharyya, A., Panchal, S., Lee, M., Madan, P., Memisevic, R.
    ICCV Workshops 2023
  4. Is end-to-end learning enough for fitness activity recognition?
    Mercier, A., Berger, G., Panchal, S., Letsch, F., Böhm, C., Kang, N., Bax, I., Memisevic, R.
    arXiv:2305.08191
  5. Unleashing the creative mind: Language model as hierarchical policy for improved exploration on challenging problem solving
    Ling, Z., Fang, Y., Li, X., Mu, T., Lee, M., Pourreza, R., Memisevic, R., Su, H.
    arXiv:2311.00694

2022

  1. Metaphors we learn by
    Memisevic, R.
    arXiv:2211.06441
  2. Decision making as language generation
    Memisevic, R., Panchal, S., Lee, M.
    NeurIPS Workshop on Foundation Models for Decision Making

2020

  1. Putting the skeleton back in the closet
    Mercier, A., Berger, G., Panchal, S., Letsch, F., Boehm, C., Bax, I., Memisevic, R.
    Medium (TwentyBN blog)  ·  a later version of this work appeared as “Is end-to-end learning enough for fitness activity recognition?” (2023)

2019

  1. The Jester dataset: A large-scale video dataset of human gestures
    Materzynska, J., Berger, G., Bax, I., Memisevic, R.
    International Conference on Computer Vision (ICCV 2019), HANDS Workshop

2018

  1. The more fine-grained, the better for transfer learning
    Mahdisoltani, F., Berger, G., Gharbieh, W., Memisevic, R., Fleet, D.
    NeurIPS 2018 Workshop on Modeling and Decision-Making in the Spatiotemporal Domain
  2. Evaluating visual “common sense” using fine-grained classification and captioning tasks
    Goyal, R., Mahdisoltani, F., Berger, G., Gharbieh, W., Bax, I., Memisevic, R.
    ICLR 2018 Workshop
  3. Hierarchical video understanding
    Mahdisoltani, F., Memisevic, R., Fleet, D.
    ECCV 2018 Workshops

2017

  1. The “something something” video database for learning and evaluating visual common sense
    Goyal, R., Kahou, S. E., Michalski, V., …, Memisevic, R.
    International Conference on Computer Vision (ICCV 2017)
  2. Incorporating long-range consistency in CNN-based texture generation
    Berger, G., Memisevic, R.
    International Conference on Learning Representations (ICLR 2017)
  3. RATM: Recurrent attentive tracking model
    Ebrahimi Kahou, S., Michalski, V., Memisevic, R., Pal, C., Vincent, P.
    CVPR 2017 BNMR Workshop
  4. Denoising criterion for variational auto-encoding framework
    Im, D. J., Ahn, S., Memisevic, R., Bengio, Y.
    AAAI Conference on Artificial Intelligence (AAAI 2017)

2016

  1. Neural networks with few multiplications
    Lin, Z., Courbariaux, M., Memisevic, R., Bengio, Y.
    International Conference on Learning Representations (ICLR 2016)
  2. Regularizing RNNs by stabilizing activations
    Krueger, D., Memisevic, R.
    International Conference on Learning Representations (ICLR 2016)
  3. Conservativeness of untied auto-encoders
    Im, D. J., Belghazi, M., Memisevic, R.
    AAAI Conference on Artificial Intelligence (AAAI 2016)
  4. Deep learning vector quantization
    de Vries, H., Memisevic, R., Courville, A.
    European Symposium on Artificial Neural Networks (ESANN 2016)
  5. Architectural complexity measures of recurrent neural networks
    Zhang, S., Wu, Y., Che, T., Lin, Z., Memisevic, R., Salakhutdinov, R., Bengio, Y.
    Neural Information Processing Systems (NIPS 2016)
  6. Theano: A Python framework for fast computation of mathematical expressions
    The Theano Development Team (incl. Memisevic, R. among 100+ co-authors)
    arXiv:1605.02688
  7. EmoNets: Multimodal deep learning approaches for emotion recognition in video
    Ebrahimi Kahou, S., Bouthillier, X., Lamblin, P., Gulcehre, C., Michalski, V., Konda, K., …, Memisevic, R., …, Bengio, Y.
    Journal on Multimodal User Interfaces, 10(2) · built on the EmotiW 2013 Grand Challenge-winning system
  8. Generating images with recurrent adversarial networks
    Im, D. J., Kim, C. D., Jiang, H., Memisevic, R.
    arXiv:1602.05110
  9. How far can we go without convolution: Improving fully-connected networks
    Lin, Z., Memisevic, R., Konda, K.
    ICLR 2016 Workshop
  10. Generative adversarial metric
    Im, D. J., Kim, C. D., Jiang, H., Memisevic, R.
    ICLR 2016 Workshop

2015

  1. The potential energy of an autoencoder
    Kamyshanska, H., Memisevic, R.
    IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI)
  2. Zero-bias autoencoders and the benefits of co-adapting features
    Konda, K., Memisevic, R., Krueger, D.
    International Conference on Learning Representations (ICLR 2015)
  3. Learning visual odometry with a convolutional network
    Konda, K., Memisevic, R.
    International Conference on Computer Vision Theory and Applications (VISAPP 2015)
  4. On using very large target vocabulary for neural machine translation
    Jean, S., Cho, K., Memisevic, R., Bengio, Y.
    Annual Meeting of the Association for Computational Linguistics (ACL 2015)
  5. Real-time activity recognition via deep learning of motion features
    Konda, K., Chandrashekhariah, P., Memisevic, R., Triesch, J.
    European Symposium on Artificial Neural Networks (ESANN 2015)
  6. Recurrent neural networks for emotion recognition in video
    Ebrahimi Kahou, S., Michalski, V., Konda, K., Memisevic, R., Pal, C.
    ACM International Conference on Multimodal Interaction (ICMI 2015)
  7. Montreal neural machine translation systems for WMT'15
    Jean, S., Firat, O., Cho, K., Memisevic, R., Bengio, Y.
    Tenth Workshop on Statistical Machine Translation (WMT 2015)
  8. Dropout as data augmentation
    Bouthillier, X., Konda, K., Vincent, P., Memisevic, R.
    arXiv:1506.08700

2014

  1. Modeling deep temporal dependencies with recurrent “grammar cells”
    Michalski, V., Memisevic, R., Konda, K.
    Neural Information Processing Systems (NIPS 2014)
  2. Learning to encode motion using spatio-temporal synchrony
    Konda, K., Memisevic, R., Michalski, V.
    International Conference on Learning Representations (ICLR 2014)
  3. A unified approach to learning depth and motion features
    Konda, K., Memisevic, R.
    Indian Conference on Computer Vision, Graphics and Image Processing (ICVGIP 2014)

2013

  1. Learning to relate images
    Memisevic, R.
    IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI), special issue on Learning Deep Architectures
  2. On autoencoder scoring
    Kamyshanska, H., Memisevic, R.
    International Conference on Machine Learning (ICML 2013)
  3. Learning invariant features by harnessing the aperture problem
    Memisevic, R., Exarchakis, G.
    International Conference on Machine Learning (ICML 2013)
  4. Feature grouping from spatially constrained multiplicative interaction
    Bauer, F., Memisevic, R.
    International Conference on Learning Representations (ICLR 2013)
  5. Combining modality-specific deep neural networks for emotion recognition in video
    Ebrahimi Kahou, S., Pal, C., Bouthillier, X., …, Memisevic, R., Vincent, P., Courville, A., Bengio, Y.
    ICMI 2013 — winner, EmotiW Grand Challenge

2012

  1. Shared kernel information embedding for discriminative inference
    Memisevic, R., Sigal, L., Fleet, D.
    IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI)
  2. On multi-view feature learning
    Memisevic, R.
    International Conference on Machine Learning (ICML 2012) · oral
  3. On spatio-temporal feature learning
    Memisevic, R.
    The Learning Workshop, Snowbird

2011

  1. Gradient-based learning of higher-order image features
    Memisevic, R.
    International Conference on Computer Vision (ICCV 2011)
  2. Modeling the joint density of two images under a variety of transformations
    Susskind, J., Memisevic, R., Hinton, G., Pollefeys, M.
    Computer Vision and Pattern Recognition (CVPR 2011)
  3. On spatio-temporal sparse coding: Analysis and an algorithm
    Memisevic, R.
    NIPS Workshop on Deep Learning and Unsupervised Feature Learning
  4. Stereopsis via deep learning
    Memisevic, R., Conrad, C.
    NIPS Workshop on Deep Learning and Unsupervised Feature Learning

2010

  1. Learning to represent spatial transformations with factored higher-order Boltzmann machines
    Memisevic, R., Hinton, G.
    Neural Computation, 22(6): 1473–1492 · cover article
  2. Gated softmax classification
    Memisevic, R., Zach, C., Hinton, G., Pollefeys, M.
    Neural Information Processing Systems (NIPS 2010)

2009

  1. Shared kernel information embedding for discriminative inference
    Sigal, L., Memisevic, R., Fleet, D.
    Computer Vision and Pattern Recognition (CVPR 2009)

2007

  1. Unsupervised learning of image transformations
    Memisevic, R., Hinton, G. E.
    Computer Vision and Pattern Recognition (CVPR 2007)
  2. Learning to solve QBF
    Samulowitz, H., Memisevic, R.
    AAAI Conference on Artificial Intelligence (AAAI 2007)

2006

  1. Kernel information embeddings
    Memisevic, R.
    International Conference on Machine Learning (ICML 2006)
  2. Propagating errors and beliefs for large scale nonlinear structure prediction
    Memisevic, R.
    North East Student Colloquium on Artificial Intelligence (NESCAI), Ithaca, NY
  3. An introduction to structured discriminative learning
    Memisevic, R.
    Technical report, University of Toronto

2005

  1. Principal surfaces from unsupervised kernel regression
    Meinicke, P., Klanke, S., Memisevic, R., Ritter, H.
    IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI), 27(9): 1379–1391
  2. Multiple relational embedding
    Memisevic, R., Hinton, G. E.
    Neural Information Processing Systems (NIPS 2005)
  3. Improving dimensionality reduction with spectral gradient descent
    Memisevic, R., Hinton, G. E.
    Neural Networks, 18, 702–710
  4. Embedding via clustering: Using spectral information to guide dimensionality reduction
    Memisevic, R., Hinton, G. E.
    International Joint Conference on Neural Networks (IJCNN 2005)
  5. Dual optimization of conditional probability models
    Memisevic, R.
    NIPS Workshop on Kernel Methods for Structured Domains
  6. Exploring the regularization path for adaptive Gaussian kernel SVMs
    Memisevic, R., Srebro, N., Roweis, S.
    NIPS Workshop on the Accuracy–Regularization Frontier

Theses

  1. Non-linear latent factor models for revealing structure in high-dimensional data
    Memisevic, R.
    PhD thesis, Department of Computer Science, University of Toronto, 2008 · advisor: Geoffrey Hinton
  2. Unsupervised kernel regression for nonlinear dimensionality reduction
    Memisevic, R.
    Master's thesis, Department of Computer Science, Bielefeld University, 2003 · advisors: Helge Ritter and Peter Meinicke

See citation counts and additional papers on Google Scholar →