Pith. sign in

REVIEW 3 major objections 5 minor 38 references

BILLNET: A Binarized Conv3D-LSTM Network with Logic-gated residual architecture for hardware-efficient video inference

T0 review · 3 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read A fully quantized Conv3D-LSTM called BILLNET reaches 83.97% on the Jester gesture benchmark with only 1.01 Mb of weights, using bitwise and bit-count operations alone.

desk verdict A genuinely useful binarized Conv3D-LSTM architecture whose weight-quantized version is solid, but the fully-quantized S5 claim is only conditionally supported by one hand-tuned recipe on one dataset. read the letter →

arxiv 2501.14495 v1 pith:I56T4N4X submitted 2025-01-24 cs.CV cs.AR

Van Thien Nguyen , William Guicquero , Gilles Sicard This is my paper · ORCID
classification cs.CVcs.AR
keywords binarizedneuralnetworksConv3DfactorizationLSTMquantizationvideoactionrecognitionhardware-efficientinferenceMUX-ORresidualbitwiseoperationsquantization-awaretraining
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

This paper claims that a video-recognition network built from 3D convolutions and a long short-term memory layer can be quantized so aggressively that its inference arithmetic reduces to bitwise and bit-count operations without losing practical accuracy. The proposed architecture, BILLNET, pairs a factorized Conv3D block with a MUX-OR residual connection, and a five-stage training procedure extends quantization all the way into the LSTM gates. On the Jester hand-gesture dataset, the fully quantized model reports 83.97% top-1 accuracy with a 1.01 Mb weight memory and 6.34 GBOPs, which the paper compares against compact 3D baselines at over 100 times larger memory and over 20 times higher bit-operation cost. The motivating claim is that fully quantized recurrent video models can be deployed on hardware that only performs bitwise and bit-count operations.

What carries the argument

The load-bearing mechanism is the combination of the MUX-OR residual block and the quantized LSTM recipe. The 2-MUX layer takes two binary feature maps, a fresh Heaviside output and an OR-gated skip path, and selects between them channel-wise using a binary control signal derived from thresholded global average pooling; this gives a parameter-free attention-like effect while keeping all data in $\{0,1\}$. The quantization identity $\mathrm{Clip}(x_1+x_2)=x_1\vee x_2$ is what lets the residual addition be implemented as an OR gate. On the recurrent side, the scaled sign weight initialization $SSign(w)=3\sqrt{n_i+n_o}\,\mathrm{Sign}(w)$ is the hand-chosen factor that keeps the quantized LSTM trainable, and the five-stage schedule, which ends with BitShift Normalization replacing batch-normalization scaling by a power-of-two bitshift, is what removes the last full-precision arithmetic from the deployed model.

What would settle it

A synthesis of the S5 model that requires any multiplier or floating-point unit in its data path, or a reproduction run whose S4-to-S5 accuracy drop is substantially larger than the reported 3.78 points, would settle the deployment claim.

Watch

Extended reading notes

Core claim

The paper's central claim is that a Conv3D-LSTM can be made fully quantized for video inference. It introduces a factorized convolution block that splits a 3D convolution into two pointwise convolutions with filter size $1\times1\times1$ around a grouped convolution with filter size $3\times3\times3$, and a 3D MUX-OR residual block in which a channel-wise multiplexer, steered by thresholded global average pooling, chooses between an OR-gated skip path and the current binary feature map. Because $\mathrm{Clip}(x_1+x_2)=x_1\vee x_2$, the residual and attention-like connections stay in binary arithmetic. The LSTM cells are then fully quantized: weights use the scaled sign $SSign(w)=3\sqrt{n_i+n_o}\,\mathrm{Sign}(w)$, sigmoid and tanh are replaced by Heaviside and sign, and the cell state is clipped to $\{-1,0,+1\}$, so the final model needs only bitwise and bit-count operations. After the five-stage training procedure, the fully quantized BILLNET reaches 83.97% top-1 accuracy on Jester with 1.01 Mb of weights and 6.34 GBOPs.

Load-bearing premise

The load-bearing premise is that the hand-tuned LSTM weight scaling factor and the approximate gradient rule keep the fully quantized recurrent layer trainable, and the paper itself records a 3.78% accuracy drop when the LSTM activations are quantized, naming that step as future work.

Editorial extensions

If this is right

  • The fully quantized S5 model can be mapped to hardware that supports only bitwise and bit-count operations, so an FPGA or ASIC implementation needs no multiply-accumulate units in the data path.
  • Removing batch normalization through power-of-two BitShift Normalization cuts the measured cost from 8.53 to 6.39 GBOPs, making that stage necessary to realize the hardware gain.
  • The five-stage training procedure provides a reusable template for quantizing other recurrent video models, with each stage isolating a single source of quantization error.
  • On Jester, the weight-quantized BILLNET already beats the 3D-MobileNet V1 baseline in accuracy while using roughly 1% of its memory and 17% of its computation, so the efficiency gain does not wait for the final LSTM quantization.

Reading between the lines

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

  • A natural test the paper leaves implicit is whether the scaled-sign LSTM recipe transfers to larger hidden state sizes and longer sequences; the reported 3.78-point S4-to-S5 drop suggests the recipe is the fragile part.
  • The MUX-OR residual can be read as a binary, parameter-free channel attention mechanism; probing it in deeper 3D networks would show whether the attention-like effect generalizes beyond this architecture.
  • Since the model keeps the temporal dimension intact after the first convolution, the same design may also suit low-latency streaming video classification on edge sensors, a setting the paper does not evaluate.
  • If the gradual sharpening of LSTM activations mentioned as future work recovers most of the S5 drop, the fully quantized model would become competitive with the S4 accuracy level while keeping bitwise-only arithmetic.
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 proposes BILLNET, a binarized Conv3D-LSTM for video gesture recognition, with a factorized Conv3D block, a MUX-OR residual connection that keeps activations binary, and a five-stage training schedule that progressively quantizes weights, activations, batch normalization, and LSTM hidden-state activations. The central reported result is on Jester: a fully quantized model with a 1.01 Mb model size and 6.34 GBOPs at 83.97% top-1 accuracy (S5), and a weight-quantized variant (S2) with 1.01 Mb, 24.54 GBOPs, and 91.64%.

Significance. If the LSTM quantization recipe proves transferable, the paper would be one of the first demonstrations of a fully quantized Conv3D-LSTM video model, and the achieved memory and computation reductions are practically relevant for edge inference: roughly a 30-fold model-size reduction and one-to-two-order-of-magnitude GBOP reduction relative to the listed full-precision compact 3D CNNs. The core Clip-as-OR arithmetic is correct, the multi-stage training idea is coherent, and the bit-shift batch-normalization simplification is a useful hardware-oriented step. The main caveats are that the fully quantized deployment claim rests on a single hand-tuned LSTM quantization recipe with no sensitivity analysis, and that S5 still uses ternary values and integer additions, so the "only bit-wise and bit-count operations" wording overstates what is demonstrated.

major comments (3)
  1. [V-B / Conclusion] The conclusion and Section V-B describe S5 as deployable on hardware supporting "only bit-wise and bit-count operations," but S5 is not purely binary: Section III-C2 keeps h_t and c_t in {-1,0,1}, Eq. (7) sums ternary products and Eq. (1) clips the result, and Eq. (10) keeps ternary weights in the last Dense layer. No decomposition of these operations into bitwise plus bit-count primitives is provided, so the central deployment claim is stronger than the reported model supports. Please either provide an explicit ternary-to-bitwise decomposition or qualify the claim.
  2. [III-C2, Eq. (9), Table I] The fully quantized LSTM recipe is introduced with a hand-chosen scaling factor in Eq. (9) and straight-through gradient estimates, and Table I reports a single configuration (g=4, n=64, m=32) with no variance information. The S4-to-S5 drop is 3.78% (87.75 to 83.97), and the conclusion lists revising the final training stage as future work. Since the paper's claim that a fully quantized Conv3D-LSTM can be trained at all rests on this recipe, the absence of sensitivity analysis, ablations on the scaling factor, and results on other datasets or state sizes leaves the generality of the central deployment claim unsupported.
  3. [Abstract/Conclusion vs. Table I] The abstract and conclusion use "high accuracy" and "on-par accuracy" without tying them to a specific training stage. Against the full-precision baselines in Table I, S5 (83.97%) is 6.8 to 9.4 points below all four listed baselines, while S2 (91.64%) is within 1.7 points of three of the four. As written, a reader can reasonably take the on-par claim to cover the fully quantized model, which the data contradict. The accuracy claim should be qualified per training stage, and ideally the comparison should also include quantized video models from the related work so that the fully quantized claim is benchmarked against relevant prior art.
minor comments (5)
  1. [Eqs. (9)-(10)] The notation "3√ni+no" and "1√4m" is ambiguous: it is not clear whether the former is three times the square root of (n_i+n_o) or a cube root, and whether the latter is the reciprocal of sqrt(4m). Please use unambiguous fraction and radical notation.
  2. [III-C2] There is a typo, "Clipped Idendity," and the paragraph would benefit from stating explicitly in the architecture overview that the quantized LSTM outputs h_t are ternary rather than binary.
  3. [IV, Eq. (11)-(12)] The claim that the equivalent batch-normalization scaling factors are "always positive" is not justified, since the learned γ can in principle be negative, and the removal of the offset β is asserted rather than analyzed; the small S3-to-S4 accuracy change is encouraging, but the argument needs an empirical or formal justification.
  4. [III-B / Table I] The symbol m is used both as a width multiplier (m=n/2 in Table I) and as the TGAP threshold parameter in Section III-B; please rename one of them to avoid confusion.
  5. [Table I] Please report variance or multiple seeds; without them, the 0.6-1.7 point differences between S2 and the full-precision baselines may not be distinguishable from training noise.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: BILLNET is an empirical architecture paper; its accuracy numbers are measured on Jester and compared against external baselines, while its hand-set design choices (Eq. 9 scaling, TGAP threshold) are inputs to training, not fitted targets renamed as predictions.

full rationale

The paper's central claims are supported by experiments on the Jester dataset, with model size and GBOPs computed from the implemented architecture and compared against external baselines from [9]. The scaling factor SSign(w) = 3*sqrt(ni+no) Sign(w) in Eq. (9), the TGAP threshold T(x) = 1{x>0.5m}, and the straight-through gradient estimates are introduced as design choices before training; they are not fit to the reported 83.97% accuracy, so no reported result reduces to a fitted parameter. The multi-stage procedure (S1 through S5) is an empirical training recipe, and the S4-to-S5 accuracy drop from 87.75% to 83.97% is presented as a measured degradation, with the authors explicitly listing the improvement of the last training stage as future work. There are no load-bearing self-citations, and no derivation in the paper is equivalent by construction to its own inputs. The main caveats — that the LSTM quantization recipe is hand-tuned and validated on a single dataset, and that S5 still contains ternary LSTM outputs and ternary last-dense weights, so the phrase 'only bit-wise and bit-count operations' is slightly overstated — are generalization/correctness risks, not circularity.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claim rests on design constants chosen by hand, the validity of straight-through gradient training for recurrent binarized layers, and the fairness of the baseline comparison. No new physical or mathematical entities are postulated; MUX-OR and QLSTM are architectural constructions, not independent entities.

free parameters (3)
  • LSTM weight scaling factor constant = 3 * (ni+no)^(1/3)
    Eq. (9); hand-chosen as a compromise between gradient propagation and matching bipolar activation distributions, with no derivation or search procedure reported.
  • TGAP threshold factor m = 0.5 * m, with m set to the layer-wise max AP output in full precision and then replaced by 1 in the quantized model
    Used in the MUX control signal; the threshold equals half the spatial resolution, 24 for a 6x8 feature map, and is chosen by hand.
  • Dense layer ternarization scale = 1/sqrt(4m)
    Eq. (10); the authors note it is removed in hardware because it does not affect Sign, Heaviside, or Argmax, so it is not load-bearing.
assumptions (4)
  • domain assumption Straight-through estimator gradients are adequate for training binarized Conv3D and LSTM layers.
    Invoked in Section III-B and III-C1; the paper reports a significant S5 accuracy drop, indicating this assumption is the fragile part.
  • standard math Clip(x1+x2) equals the logical OR of binarized feature maps, so the MUX-OR gate preserves binary semantics.
    Eq. (1) and Section III-B; correct for binary inputs in {0,1}, but relies on feature maps being exactly binary after Heaviside.
  • standard math Batch normalization can be replaced by positive bitshift scaling without changing the output after Heaviside.
    Section IV stage 4; true because positive scale factors preserve sign, but depends on the trained BN scale being positive.
  • domain assumption Jester Top-1 accuracy and the BOPs metric computed from [9]'s public code are a fair basis for comparing efficiency.
    Section V; the paper does not specify whether the baseline evaluation protocol, including sampling, resolution, and clip length, matches BILLNET exactly.

how reviews work

0 comments
Cite this review

Pith. "Pith review of BILLNET: A Binarized Conv3D-LSTM Network with Logic-gated residual architecture for hardware-efficient video inference." pith.science (2026). https://pith.science/paper/I56T4N4X

@misc{pith2026250114495,
  author       = {Pith},
  title        = {Pith review of: BILLNET: A Binarized Conv3D-LSTM Network with Logic-gated residual architecture for hardware-efficient video inference},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/I56T4N4X}},
  note         = {Machine review of arXiv:2501.14495}
}
read the original abstract

Long Short-Term Memory (LSTM) and 3D convolution (Conv3D) show impressive results for many video-based applications but require large memory and intensive computing. Motivated by recent works on hardware-algorithmic co-design towards efficient inference, we propose a compact binarized Conv3D-LSTM model architecture called BILLNET, compatible with a highly resource-constrained hardware. Firstly, BILLNET proposes to factorize the costly standard Conv3D by two pointwise convolutions with a grouped convolution in-between. Secondly, BILLNET enables binarized weights and activations via a MUX-OR-gated residual architecture. Finally, to efficiently train BILLNET, we propose a multi-stage training strategy enabling to fully quantize LSTM layers. Results on Jester dataset show that our method can obtain high accuracy with extremely low memory and computational budgets compared to existing Conv3D resource-efficient models.

Figures

Figures reproduced from arXiv: 2501.14495 by the authors.

Figure 1
Figure 1. Top-level architecture description of BILLNET with [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. The operation of the channel-wise MUX gate with [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Computational graph of the proposed Quantized LSTM. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (2 more)
Figure 6
Figure 6. Figure 6: Class-temporal BILLNET output responses for a Jester [PITH_FULL_IMAGE:figures/full_fig_p005_6.png]
Figure 5
Figure 5. Figure 5: Weight memory (Mb) and computational costs (GBOPs [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

38 extracted references · 30 canonical work pages

  1. [1]

    Gate-shift networks for video action recognition,

    S. Sudhakaran, S. Escalera, and O. Lanz, “Gate-shift networks for video action recognition,” in 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , 2020, pp. 1099–1108

    work page 2020

  2. [2]

    Temporal segment networks for action recognition in videos,

    L. Wang, Y . Xiong, Z. Wang, Y . Qiao, D. Lin, X. Tang, and L. Van Gool, “Temporal segment networks for action recognition in videos,” IEEE Transactions on Pattern Analysis and Machine Intelligence , vol. 41, no. 11, pp. 2740–2755, 2019

    work page 2019

  3. [3]

    Gating re- visited: Deep multi-layer RNNs that can be trained,

    M. O. Turkoglu, S. D’Aronco, J. Wegner, and K. Schindler, “Gating re- visited: Deep multi-layer RNNs that can be trained,” IEEE Transactions on Pattern Analysis and Machine Intelligence , pp. 1–1, 2021

    work page 2021

  4. [4]

    Quo vadis, action recognition? a new model and the Kinetics dataset,

    J. Carreira and A. Zisserman, “Quo vadis, action recognition? a new model and the Kinetics dataset,” in 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , 2017, pp. 4724–4733

    work page 2017

  5. [5]

    YouTube-8M: A large-scale video clas- sification benchmark,

    S. Abu-El-Haija, N. Kothari, J. Lee, A. Natsev, G. Toderici, B. Varadara- jan, and S. Vijayanarasimhan, “YouTube-8M: A large-scale video clas- sification benchmark,” ArXiv, vol. abs/1609.08675, 2016

    arXiv 2016

  6. [6]

    The Jester dataset: A large-scale video dataset of human gestures,

    J. Materzynska, G. Berger, I. Bax, and R. Memisevic, “The Jester dataset: A large-scale video dataset of human gestures,” in 2019 IEEE/CVF International Conference on Computer Vision Workshop (ICCVW), 2019, pp. 2874–2882

    work page 2019

  7. [7]

    Can spatiotemporal 3D cnns retrace the history of 2D cnns and imagenet?

    K. Hara, H. Kataoka, and Y . Satoh, “Can spatiotemporal 3D cnns retrace the history of 2D cnns and imagenet?” in 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2018, pp. 6546–6555

    work page 2018

  8. [8]

    Long Short-Term Memory,

    S. Hochreiter and J. Schmidhuber, “Long Short-Term Memory,” Neural Comput., vol. 9, no. 8, p. 1735–1780, nov 1997

    work page 1997

Show all 38 references
  1. [9]

    Resource efficient 3D convolutional neural networks,

    O. K ¨op¨ukl¨u, N. Kose, A. Gunduz, and G. Rigoll, “Resource efficient 3D convolutional neural networks,” in 2019 IEEE/CVF International Conference on Computer Vision Workshop (ICCVW) , 2019, pp. 1910– 1919

    2019

  2. [10]

    Tiny video networks,

    A. J. Piergiovanni, A. Angelova, and M. S. Ryoo, “Tiny video networks,” ArXiv, vol. abs/1910.06961, 2019

    1910 arXiv

  3. [11]

    3D CNN acceleration on FPGA using hardware-aware pruning,

    M. Sun, P. Zhao, M. Gungor, M. Pedram, M. Leeser, and X. Lin, “3D CNN acceleration on FPGA using hardware-aware pruning,” in 2020 57th ACM/IEEE Design Automation Conference (DAC) , 2020, pp. 1–6

    2020

  4. [12]

    Quantized neural networks: Training neural networks with low preci- sion weights and activations,

    I. Hubara, M. Courbariaux, D. Soudry, R. El-Yaniv, and Y . Bengio, “Quantized neural networks: Training neural networks with low preci- sion weights and activations,” J. Mach. Learn. Res. , vol. 18, no. 1, p. 6869–6898, jan 2017

    2017

  5. [13]

    Bi- narized neural networks,

    I. Hubara, M. Courbariaux, D. Soudry, R. El-Yaniv, and Y . Bengio, “Bi- narized neural networks,” in Advances in Neural Information Processing Systems (NeurIPS), 2016, pp. 4107–4115

    2016

  6. [14]

    Learned step size quantization,

    S. K. Esser, J. L. McKinstry, D. Bablani, R. Appuswamy, and D. S. Modha, “Learned step size quantization,” in 8th International Confer- ence on Learning Representations, ICLR 2020, Addis Ababa, Ethiopia, April 26-30, 2020 , 2020

    2020

  7. [15]

    Recurrent neural networks with limited numerical precision,

    J. Ott, Z. Lin, Y . Zhang, S.-C. Liu, and Y . Bengio, “Recurrent neural networks with limited numerical precision,” ArXiv, vol. abs/1611.07065, 2016

    2016 arXiv

  8. [16]

    Effective quantization approaches for recurrent neural networks,

    M. Z. Alom, A. T. Moody, N. Maruyama, B. C. Van Essen, and T. M. Taha, “Effective quantization approaches for recurrent neural networks,” in 2018 International Joint Conference on Neural Networks (IJCNN) , 2018, pp. 1–8

    2018

  9. [17]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” in 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016, pp. 770–778

    2016

  10. [18]

    Residual attention network for image classification,

    F. Wang, M. Jiang, C. Qian, S. Yang, C. Li, H. Zhang, X. Wang, and X. Tang, “Residual attention network for image classification,” in 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , 2017, pp. 6450–6458

    2017

  11. [19]

    Squeeze-and-excitation networks,

    J. Hu, L. Shen, S. Albanie, G. Sun, and E. Wu, “Squeeze-and-excitation networks,” IEEE Transactions on Pattern Analysis and Machine Intelli- gence, vol. 42, no. 8, pp. 2011–2023, 2020

    2011

  12. [20]

    Learning spatio-temporal features with 3D residual networks for action recognition,

    K. Hara, H. Kataoka, and Y . Satoh, “Learning spatio-temporal features with 3D residual networks for action recognition,” in 2017 IEEE In- ternational Conference on Computer Vision Workshops (ICCVW), 2017, pp. 3154–3160

    2017

  13. [21]

    Revisiting 3D resnets for video recognition,

    X. Du, Y . Li, Y . Cui, R. Qian, J. Li, and I. Bello, “Revisiting 3D resnets for video recognition,” ArXiv, vol. abs/2109.01696, 2021

    2021 arXiv

  14. [22]

    Spatio- temporal attention networks for action recognition and detection,

    J. Li, X. Liu, W. Zhang, M. Zhang, J. Song, and N. Sebe, “Spatio- temporal attention networks for action recognition and detection,” IEEE Transactions on Multimedia , vol. 22, no. 11, pp. 2990–3001, 2020

    2020

  15. [23]

    A closer look at spatiotemporal convolutions for action recognition,

    D. Tran, H. Wang, L. Torresani, J. Ray, Y . LeCun, and M. Paluri, “A closer look at spatiotemporal convolutions for action recognition,” in 2018 IEEE/CVF Conference on Computer Vision and Pattern Recogni- tion, 2018, pp. 6450–6459

    2018

  16. [24]

    Rethinking spatio- temporal feature learning: Speed-accuracy trade-offs in video classifica- tion,

    S. Xie, C. Sun, J. Huang, Z. Tu, and K. Murphy, “Rethinking spatio- temporal feature learning: Speed-accuracy trade-offs in video classifica- tion,” in Computer Vision - ECCV 2018 - 15th European Conference, Munich, Germany, September 8-14, 2018, Proceedings, Part XV, V . Fer-...

    2018

  17. [25]

    Learning spatio-temporal representation with pseudo-3D residual networks,

    Z. Qiu, T. Yao, and T. Mei, “Learning spatio-temporal representation with pseudo-3D residual networks,” in 2017 IEEE International Con- ference on Computer Vision (ICCV) , 2017, pp. 5534–5542

    2017

  18. [26]

    TSM: Temporal shift module for efficient video understanding,

    J. Lin, C. Gan, and S. Han, “TSM: Temporal shift module for efficient video understanding,” in 2019 IEEE/CVF International Conference on Computer Vision (ICCV) , 2019, pp. 7082–7092

    2019

  19. [28]

    Shufflenet: An extremely efficient convolutional neural network for mobile devices,

    X. Zhang, X. Zhou, M. Lin, and J. Sun, “Shufflenet: An extremely efficient convolutional neural network for mobile devices,” in 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2018, pp. 6848–6856

    2018

  20. [29]

    Dynamic network quantization for efficient video inference,

    X. Sun, R. Panda, C.-F. R. Chen, A. Oliva, R. Feris, and K. Saenko, “Dynamic network quantization for efficient video inference,” in Pro- ceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), October 2021, pp. 7375–7385

    2021

  21. [30]

    Efficient binary 3D convo- lutional neural network and hardware accelerator,

    G. Li, M. Zhang, Q. Zhang, and Z. Lin, “Efficient binary 3D convo- lutional neural network and hardware accelerator,” J. Real Time Image Process., vol. 19, no. 1, pp. 61–71, 2022

    2022

  22. [31]

    4-bit quantization of LSTM-based speech recognition models,

    A. Fasoli, C. Chen, M. J. Serrano, X. Sun, N. Wang, S. Venkataramani, G. Saon, X. Cui, B. Kingsbury, W. Zhang, Z. T ¨uske, and K. Gopalakr- ishnan, “4-bit quantization of LSTM-based speech recognition models,” CoRR, vol. abs/2108.12074, 2021

    2021 arXiv

  23. [32]

    HitNet: Hybrid ternary recurrent neural network,

    P. Wang, X. Xie, L. Deng, G. Li, D. Wang, and Y . Xie, “HitNet: Hybrid ternary recurrent neural network,” in Advances in Neural Information Processing Systems (NeurIPS) , vol. 31, 2018

    2018

  24. [33]

    DC-LSTM: Deep com- pressed LSTM with low bit-width and structured matrices,

    G. Nan, C. Wang, W. Liu, and F. Lombardi, “DC-LSTM: Deep com- pressed LSTM with low bit-width and structured matrices,” in 2020 IEEE International Symposium on Circuits and Systems (ISCAS) , 2020, pp. 1–5

    2020

  25. [34]

    Estimating or propagat- ing gradients through stochastic neurons for conditional computation,

    Y . Bengio, N. L ´eonard, and A. Courville, “Estimating or propagat- ing gradients through stochastic neurons for conditional computation,” arXiv:1308.3432 [cs], Aug. 2013

    2013 arXiv

  26. [35]

    Batch normalization: Accelerating deep network training by reducing internal covariate shift,

    S. Ioffe and C. Szegedy, “Batch normalization: Accelerating deep network training by reducing internal covariate shift,” ArXiv, vol. abs/1502.03167, 2015

    2015 arXiv

  27. [36]

    Ternary weight networks,

    F. Li and B. Liu, “Ternary weight networks,” CoRR, vol. abs/1605.04711, 2016. [Online]. Available: http://arxiv.org/abs/1605. 04711

    2016 arXiv

  28. [37]

    Adam: A method for stochastic optimiza- tion,

    D. P. Kingma and J. Ba, “Adam: A method for stochastic optimiza- tion,” in 3rd International Conference on Learning Representations, ICLR 2015, San Diego, CA, USA, May 7-9, 2015, Conference Track Proceedings, Y . Bengio and Y . LeCun, Eds., 2015

    2015

  29. [38]

    Differentiable joint pruning and quantization for hardware efficiency,

    Y . Wang, Y . Lu, and T. Blankevoort, “Differentiable joint pruning and quantization for hardware efficiency,” inComputer Vision - ECCV 2020 - 16th European Conference, Glasgow, UK, August 23-28, 2020, Proceed- ings, Part XXIX , ser. Lecture Notes in Computer Science, A. Vedal...

    2020

  30. [2017]

    Available: http://arxiv.org/abs/1704.04861

    [Online]. Available: http://arxiv.org/abs/1704.04861

    arXiv

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.