BRN Discussion Ongoing
- Thread starter TechGirl
- Start date
Legendre gets a mention in:
TENNs-PLEIADES: Building Temporal Kernels with Orthogonal Polynomials
Yan Ru Pei, Olivier Coenenhttps://arxiv.org/html/2405.12179v3
Abstract
We introduce a neural network named PLEIADES (PoLynomial Expansion In Adaptive Distributed Event-based Systems), belonging to the TENNs (Temporal Neural Networks) architecture. We focus on interfacing these networks with event-based data to perform online spatiotemporal classification and detection with low latency. By virtue of using structured temporal kernels and event-based data, we have the freedom to vary the sample rate of the data along with the discretization step-size of the network without additional finetuning. We experimented with three event-based benchmarks and obtained state-of-the-art results on all three by large margins with significantly smaller memory and compute costs. We achieved: 1) 99.59% accuracy with 192K parameters on the DVS128 hand gesture recognition dataset and 100% with a small additional output filter; 2) 99.58% test accuracy with 277K parameters on the AIS 2024 eye tracking challenge; and 3) 0.556 mAP with 576k parameters on the PROPHESEE 1 Megapixel Automotive Detection Dataset.1Introduction
Temporal convolutional networks (TCNs) [18] have been a staple for processing time series data from speech enhancement [22] to action segmentation [17]. However, in most cases, the temporal kernel is very short (usually size of 3), making it difficult for the network to capture long-range temporal correlations. The temporal kernels are intentionally kept short, because keeping a long temporal kernel with a large number of trainable kernel values usually leads to unstable training. In addition, we require a large amount of memory for storing the weights during inference. One popular solution for this has been to parameterize the temporal kernel function with a simple multilayer perceptron (MLP), which promotes stability [28] and more compressed parameters, but it often increases the computational load considerably.Here, we introduce a method of parameterization of temporal kernels, named PLEIADES (PoLynomial Expansion In Adaptive Distributed Event-based Systems), that can in many cases reduce the memory and computational costs compared to explicit convolutions. The design is fairly modular, and can be used as a drop-in replacement for any 1D-like convolutional layers, allowing them to perform long temporal convolutions effectively. In fact, we augment a previously proposed (1+2)D causal spatiotemporal network [23] by replacing its temporal kernels with this new polynomial parameterization. This new network architecture serves as the backbone for a wide range of online spatiotemporal tasks ranging from action recognition to object detection.This network belong to a broader class of networks named Temporal Neural Networks (TENNs) developed by Brainchip Inc.
...
The seminal work proposing a memory encoding using orthogonal Legendre polynomials in a recurrent state-space model is the Legendre Memory Unit (LMU) [33], where Legendre polynomials (a special case of Jacobi polynomials) are used. The HiPPO formalism [11] then generalized this to other orthogonal functions including Chebyshev polynomials, Laguerre polynomials, and Fourier modes. Later, this sparked a cornucopia of works interfacing with deep state space models including S4 [12], H3 [2], and Mamba [10], achieving impressive results on a wide range of tasks from audio generation to language modeling. There are several common themes among these networks that PLEIADES differ from. First, these models typically only interface with 1D temporal data, and usually try to flatten high dimensional data into 1D data before processing [12, 37], with some exceptions [21]. Second, instead of explicitly performing finite-window temporal convolutions, a running approximation of the effects of such convolutions are performed, essentially yielding a system with infinite impulse responses where the effective polynomial structures are distorted [31, 11]. And in the more recent works, the polynomial structures are tenuously used only for initialization, but then made fully trainable. Finally, these networks mostly use an underlying depthwise structure [14] for long convolutions, which may limit the network capacity, albeit reducing the compute requirement of the network.
[33]↑Aaron Voelker, Ivana Kajić, and Chris Eliasmith.Legendre Memory Units: Continuous-time representation in recurrent neural networks.Advances in neural information processing systems, 32, 2019. [Uni of Waterloo]
Pom down under
Top 20
Dallas
Regular
The Quiet Revolution in Edge AI Hardware | Dinesh Tyagi posted on the topic | LinkedIn
𝗘𝗱𝗴𝗲 𝗔𝗜 𝗖𝗵𝗶𝗽𝘀: 𝗧𝗵𝗲 𝗤𝘂𝗶𝗲𝘁 𝗥𝗲𝘃𝗼𝗹𝘂𝘁𝗶𝗼𝗻 𝗶𝗻 𝗦𝗲𝗺𝗶𝗰𝗼𝗻𝗱𝘂𝗰𝘁𝗼𝗿𝘀 As the semiconductor world chases ever-larger datacenter chips to power LLMs and generative AI, a quiet but equally transformative shift is unfolding: AI at the edge. According to a recent report by MarketsandMarkets, the edge AI hardware...
Pom down under
Top 20
Maybe we should turn the BRN thread into a dating site since so many of us are getting ex-communicated by our significant others.
I‘ll start the ball rolling.
I‘m a fun-loving lass with a quirky sense of humour and an even quirkier sense of fashion. I like to practice taekwondo in my spare time and to watch documentaries about true crime and evil psychopaths on Netflix, as well as eating tubs of Connoisseur ice-cream. I’m currently learning how to play the maracas and the bugle. I have 10 cats and a blue-tongued lizard called Gertrude.
I don’t like chewing sounds, so if you’re one of those people who can eat with your mouth closed, then please feel free to call me.
Maybe we should turn the BRN thread into a dating site since so many of us are getting ex-communicated by our significant others.
I‘ll start the ball rolling.
I‘m a fun-loving lass with a quirky sense of humour and an even quirkier sense of fashion. I like to practice taekwondo in my spare time and to watch documentaries about true crime and evil psychopaths on Netflix, as well as eating tubs of Connoisseur ice-cream. I’m currently learning how to play the maracas and the bugle
I don’t like chewing sounds, so if you’re one of those people who can eat with your mouth closed, then please feel free to call me.
The Quiet Revolution in Edge AI Hardware | Dinesh Tyagi posted on the topic | LinkedIn
𝗘𝗱𝗴𝗲 𝗔𝗜 𝗖𝗵𝗶𝗽𝘀: 𝗧𝗵𝗲 𝗤𝘂𝗶𝗲𝘁 𝗥𝗲𝘃𝗼𝗹𝘂𝘁𝗶𝗼𝗻 𝗶𝗻 𝗦𝗲𝗺𝗶𝗰𝗼𝗻𝗱𝘂𝗰𝘁𝗼𝗿𝘀 As the semiconductor world chases ever-larger datacenter chips to power LLMs and generative AI, a quiet but equally transformative shift is unfolding: AI at the edge. According to a recent report by MarketsandMarkets, the edge AI hardware...www.linkedin.com
Ladies and gentlemen, we have a Karen on board. Mimimimi
Last edited:
Mccabe84
Regular
Its better than the last one..
You still read his comments? Respect
In short
Sydney - 30 July 2025 - BrainChip Holdings Ltd (ASX: BRN, OTCQX: BRCHF, BCHPY) (Company), the world's first commercial producer of neuromorphic artificial intelligence technology, today provides the Quarterly Activities Report in conjunction with its Appendix 4C lodged for the quarter ending 30 June 2025.
Key Highlights
Redomicile Update
On 27 February 2025, the Company announced it was evaluating the possibility of redomiciling to an alternative stock exchange with a focus on the US. Post an extensive review that included input and advice from a range of experts, including foreign and domestic legal advisors, investment banks and feedback from shareholders, the Board has made the decision that shareholder value is best achieved by remaining listed on the ASX.
BrainChip remains committed to the ASX listing and ensuring that the Company continues its path to commercial success. The Board acknowledges and appreciates the ongoing commitment of shareholders. This sustained support is instrumental to the Company's progress and underpins its pursuit of long-term growth.
Sydney - 30 July 2025 - BrainChip Holdings Ltd (ASX: BRN, OTCQX: BRCHF, BCHPY) (Company), the world's first commercial producer of neuromorphic artificial intelligence technology, today provides the Quarterly Activities Report in conjunction with its Appendix 4C lodged for the quarter ending 30 June 2025.
Key Highlights
- Cash balance of US$13.5M provides sufficient capital for growth and investment in research and development of new and existing products.
- Evaluation of redomiciling Company listing has been completed with input from a range of domestic and international advisors, including legal, investment banking groups, and shareholders. After detailed evaluation and analysis, the Board made the decision that shareholder value is best achieved by remaining listed on the ASX.
- Cash inflow from customers in the current quarter of US$1.4M was higher than the prior quarter (US$0.14M).
- Total payments to suppliers and employees of US$4.4M in the current quarter were lower than the prior quarter (US$4.9M).
- Collaboration with multiple high-quality companies during the quarter further demonstrates the commercial application of BrainChip's technology.
- Continued expansion of global intellectual property portfolio, now comprising 55 issued and pending patents across the United States, Europe, and APAC regions.
Redomicile Update
On 27 February 2025, the Company announced it was evaluating the possibility of redomiciling to an alternative stock exchange with a focus on the US. Post an extensive review that included input and advice from a range of experts, including foreign and domestic legal advisors, investment banks and feedback from shareholders, the Board has made the decision that shareholder value is best achieved by remaining listed on the ASX.
BrainChip remains committed to the ASX listing and ensuring that the Company continues its path to commercial success. The Board acknowledges and appreciates the ongoing commitment of shareholders. This sustained support is instrumental to the Company's progress and underpins its pursuit of long-term growth.
TheDrooben
Pretty Pretty Pretty Pretty Good
#neuromorphiccomputing #enterpriseai #aifuture #braininspired #techtrends #cio #digitaltransformation #sustainabletech #clevel #innovation | Mohamed Alhashme, PhD
🚀 Data-Backed Prediction: Neuromorphic Computing Will Transform Energy-Efficient AI in Enterprises—Sooner Than Expected A couple months ago, I reviewed energy forecasts in a leading industry report, and the figures were daunting: AI power demands could double, straining budgets and...
Happy as Larry
Not Brainchip Akida ... but a good sign
www.linkedin.com
In a world of kilowatt GPUs, a sub-milliwatt AI chips is a novelty. And, in a world of 600-billion-parameter transformer-based LLMs, spiking neural networks (SNNs) are even more curious. In May… | Joseph B.
In a world of kilowatt GPUs, a sub-milliwatt AI chips is a novelty. And, in a world of 600-billion-parameter transformer-based LLMs, spiking neural networks (SNNs) are even more curious. In May, Innatera announced availability of Pulsar, a RISC-V chip that accelerates SNNs, a type of...
Ten times better
TheDrooben
Pretty Pretty Pretty Pretty Good
Have we finally reached the inflection point????.........revenue growing (albeit from a low base). Almost 12 months of cash left at this burn rate.....what will the next 4C bring?? Only 14 Notification regarding Unquoted Securities announcements until then.......Good 4c in my opinion
Happy as Larry
Similar threads
