From Near-Sensor to In-Sensor: A State-of-the-Art Review of Embedded AI Vision Systems.

In modern cyber-physical systems, the integration of AI into vision pipelines is now a standard practice for applications ranging from autonomous vehicles to mobile devices. Traditional AI integration often relies on cloud-based processing, which faces challenges such as data access bottlenecks, increased latency, and high power consumption. This article reviews embedded AI vision systems, examining the diverse landscape of near-sensor and in-sensor processing architectures that incorporate convolutional neural networks. We begin with a comprehensive analysis of the critical characteristics and metrics that define the performance of AI-integrated vision systems. These include sensor resolution, frame rate, data bandwidth, computational throughput, latency, power efficiency, and overall system scalability. Understanding these metrics provides a foundation for evaluating how different embedded processing architectures impact the entire vision pipeline, from image capture to AI inference. Our analysis delves into near-sensor systems that leverage dedicated hardware accelerators and commercially available components to efficiently process data close to their source, minimizing data transfer overhead and latency. These systems offer a balance between flexibility and performance, allowing for real-time processing in constrained environments. In addition, we explore in-sensor processing solutions that integrate computational capabilities directly into the sensor. This approach addresses the rigorous demand constraints of embedded applications by significantly reducing data movement and power consumption while also enabling in-sensor feature extraction, pre-processing, and CNN inference. By comparing these approaches, we identify trade-offs related to flexibility, power consumption, and computational performance. Ultimately, this article provides insights into the evolving landscape of embedded AI vision systems and suggests new research directions for the development of next-generation machine vision systems.

When the nose must remain responsive: glutathione conjugation of the mammary pheromone in the newborn rabbit.

In insects, xenobiotic-metabolizing enzymes were demonstrated to regulate pheromones inactivation, clearing them from the olfactory periphery and keeping receptors ready for stimulation renewal. Here, we investigate whether similar processes could occur in mammals, focusing on the pheromonal communication between female rabbits and their newborns. Lactating rabbits emit in their milk a volatile aldehyde, 2-methylbut-2-enal, that elicits searching-grasping in neonates; called the mammary pheromone (MP), it is critical for pups which are constrained to find nipples within the 5 min of daily nursing. For newborns, it is thus essential to remain sensitive to this odorant during the whole nursing period to display several actions of sucking. Here, we show that the MP is enzymatically conjugated to glutathione in newborn olfactory epithelium (OE), in accordance with the high mRNA expression of glutathione transferases evidenced by quantitative reverse transcription-PCR. This activity in the nose is higher than in the liver and in OE of newborns compared with weanlings (no more responsive to the pheromone). Therefore, the results pinpoint the existence of a high level of MP-glutathione conjugation activity in the OE of young rabbits, especially in the developmental window where the perceptual sensitivity toward the MP is crucial for survival.

Towards a configurable and non-hierarchical search space for NAS.

Neural Architecture Search (NAS) outperforms handcrafted Neural Network (NN) design. However, current NAS methods generally use hard-coded search spaces, and predefined hierarchical architectures. As a consequence, adapting them to a new problem can be cumbersome, and it is hard to know which of the NAS algorithm or the predefined hierarchical structure impacts performance the most. To improve flexibility, and be less reliant on expert knowledge, this paper proposes a NAS methodology in which the search space is easily customizable, and allows for full network search. NAS is performed with Gaussian Process (GP)-based Bayesian Optimization (BO) in a continuous architecture embedding space. This embedding is built upon a Wasserstein Autoencoder, regularized by both a Maximum Mean Discrepancy (MMD) penalization and a Fully Input Convex Neural Network (FICNN) latent predictor, trained to infer the parameter count of architectures. This paper first assesses the embedding's suitability for optimization by solving 2 computationally inexpensive problems: minimizing the number of parameters, and maximizing a zero-shot accuracy proxy. Then, two variants of complexity-aware NAS are performed on CIFAR-10 and STL-10, based on two different search spaces, providing competitive NN architectures with limited model sizes.

Expression of the Cerebral Olfactory Receptors Olfr110/111 and Olfr544 Is Altered During Aging and in Alzheimer's Disease-Like Mice.

A growing number of studies report the expression of olfactory receptors (ORs) in many non-chemosensory tissues and organs. However, within the brain, very few ectopic ORs are exhaustively documented. Their kinetic expression, cellular localization, and functions remain elusive. Using cDNA microarrays, quantitative PCR, and immunohistochemistry, we studied the cellular and sub-cellular localization of Olfr110/111 and Olfr544 and their timely expression in various brain areas of wild-type and transgenic Alzheimer's disease-like (5xFAD) mice. We observed that Olfr110/111 and Olfr544 proteins are mainly expressed by neurons in cortical and hippocampal regions and, to a lesser extent, by astrocytes, microglia, oligodendrocytes, and endothelial cells. In addition, both ORs are present at the cell membrane and co-expressed with the olfactory Gαolf protein, suggesting that they can be functional. Remarkably, we also found that the expression of the mRNA encoding for Olfr110/111 tends to increase with age in both the cortex and hippocampus of wild-type and transgenic mice. Moreover, Olfr110/111 transcript expression is markedly impaired in the brain of Alzheimer's disease-like mice. A different profile is noticed for Olfr544, for which an overexpression is observed only in the cortex of 9-month-old animals. In addition, in transgenic mice, olfactory receptors are observed near amyloid plaques. Altogether, our findings indicate that ORs may play a role in brain functioning, in normal and pathological conditions.

Development of the ETOC: a European test of olfactory capabilities.

A number of smell tests designed to evaluate human olfactory capabilities have been published, but none have been validated cross-culturally. The aim of this study was therefore to develop a reliable and quick olfactory test that could be used to evaluate efficiently the olfactory abilities of a European population. This test, named ETOC and based on a combination of a supra-threshold detection task and an identification task, was designed to be a cross-cultural tool that would measure the decline in olfactory performance with ageing. Two versions of the ETOC, one easy and one less easy, were used to test the olfactory performance of European citizens in three countries (France, Sweden and the Netherlands). The results indicated that neither version of the ETOC is culture-dependent, and that both give scores that well reflect the decrease in olfactory abilities with increasing age. A retest session showed that the less easy (and final) version of the ETOC is also highly reliable.

A unique method for the isolation of nasal olfactory stem cells in living rats.

Stem cells are attractive tools to develop new therapeutic strategies for a variety of disorders. While ethical and technical issues, associated with embryonic, fetal and neural stem cells, limit the translation to clinical applications, the nasal stem cells identified in the human olfactory mucosa stand as a promising candidate for stem cell-based therapies. Located in the back of the nose, this multipotent stem cell type is readily accessible in humans, a feature that makes these cells highly suitable for the development of autologous cell-based therapies. However, preclinical studies based on autologous transplantation of rodent olfactory stem cells are impeded because of the narrow opening of the nasal cavity. In this study, we report the development of a unique method permitting to quickly and safely biopsy olfactory mucosa in rats. Using this newly developed technique, rat stem cells expressing the stem cell marker Nestin were successfully isolated without requiring the sacrifice of the donor animal. As an evidence of the self-renewal capacity of the isolated cells, several millions of rat cells were amplified from a single biopsy within four weeks. Using an olfactory discrimination test, we additionally showed that this novel biopsy method does not affect the sense of smell and the learning and memory abilities of the operated animals. This study describes for the first time a methodology allowing the derivation of rat nasal cells in a way that is suitable for studying the effects of autologous transplantation of any cell type present in the olfactory mucosa in a wide variety of rat models.

Odorant metabolism catalyzed by olfactory mucosal enzymes influences peripheral olfactory responses in rats.

A large set of xenobiotic-metabolizing enzymes (XMEs), such as the cytochrome P450 monooxygenases (CYPs), esterases and transferases, are highly expressed in mammalian olfactory mucosa (OM). These enzymes are known to catalyze the biotransformation of exogenous compounds to facilitate elimination. However, the functions of these enzymes in the olfactory epithelium are not clearly understood. In addition to protecting against inhaled toxic compounds, these enzymes could also metabolize odorant molecules, and thus modify their stimulating properties or inactivate them. In the present study, we investigated the in vitro biotransformation of odorant molecules in the rat OM and assessed the impact of this metabolism on peripheral olfactory responses. Rat OM was found to efficiently metabolize quinoline, coumarin and isoamyl acetate. Quinoline and coumarin are metabolized by CYPs whereas isoamyl acetate is hydrolyzed by carboxylesterases. Electro-olfactogram (EOG) recordings revealed that the hydroxylated metabolites derived from these odorants elicited lower olfactory response amplitudes than the parent molecules. We also observed that glucurono-conjugated derivatives induced no olfactory signal. Furthermore, we demonstrated that the local application of a CYP inhibitor on rat olfactory epithelium increased EOG responses elicited by quinoline and coumarin. Similarly, the application of a carboxylesterase inhibitor increased the EOG response elicited by isoamyl acetate. This increase in EOG amplitude provoked by XME inhibitors is likely due to enhanced olfactory sensory neuron activation in response to odorant accumulation. Taken together, these findings strongly suggest that biotransformation of odorant molecules by enzymes localized to the olfactory mucosa may change the odorant's stimulating properties and may facilitate the clearance of odorants to avoid receptor saturation.

Functional evidence of multidrug resistance transporters (MDR) in rodent olfactory epithelium.

BACKGROUND: P-glycoprotein (Pgp) and multidrug resistance-associated protein (MRP1) are membrane transporter proteins which function as efflux pumps at cell membranes and are considered to exert a protective function against the entry of xenobiotics. While evidence for Pgp and MRP transporter activity is reported for olfactory tissue, their possible interaction and participation in the olfactory response has not been investigated. PRINCIPAL FINDINGS: Functional activity of putative MDR transporters was assessed by means of the fluorometric calcein acetoxymethyl ester (calcein-AM) accumulation assay on acute rat and mouse olfactory tissue slices. Calcein-AM uptake was measured as fluorescence intensity changes in the presence of Pgp or MRP specific inhibitors. Epifluorescence microscopy measured time course analysis in the olfactory epithelium revealed significant inhibitor-dependent calcein uptake in the presence of each of the selected inhibitors. Furthermore, intracellular calcein accumulation in olfactory receptor neurons was also significantly increased in the presence of either one of the Pgp or MRP inhibitors. The presence of Pgp or MRP1 encoding genes in the olfactory mucosa of rat and mouse was confirmed by RT-PCR with appropriate pairs of species-specific primers. Both transporters were expressed in both newborn and adult olfactory mucosa of both species. To assess a possible involvement of MDR transporters in the olfactory response, we examined the electrophysiological response to odorants in the presence of the selected MDR inhibitors by recording electroolfactograms (EOG). In both animal species, MRPs inhibitors induced a marked reduction of the EOG magnitude, while Pgp inhibitors had only a minor or no measurable effect. CONCLUSIONS: The findings suggest that both Pgp and MRP transporters are functional in the olfactory mucosa and in olfactory receptor neurons. Pgp and MRPs may be cellular constituents of olfactory receptor neurons and represent potential mechanisms for modulation of the olfactory response.

The lyon clinical olfactory test: validation and measurement of hyposmia and anosmia in healthy and diseased populations.

The LCOT is a self-administered test designed to assess olfactory deficits. Altogether, 525 subjects contributed to the validation. Elderly participants were well represented in this sample. In a validation study (study 1), 407 healthy and 17 anosmic volunteers between 15 and 91 years of age underwent threshold, supraliminal detection, and identification testing. Cutoff values for normosmia and hyposmia were calculated and applied in a second study in a group of patients with smell complaints and in a group of Alzheimer patients with age-matched controls. Incidence of smell deficit was estimated at 5.6% in the healthy population of study 1, and at 16% in the elderly control group of study 2. Assessment of the ability of each subtest to discriminate between groups showed that LCOT is relevant to differentiating between perception and identification deficits and between Alzheimer's and hyposmic patients.

Peripheral olfactory sensitivity in rodents after treatment with docetaxel.

OBJECTIVES/HYPOTHESIS: Clinical studies have documented that cytotoxic chemotherapy is often associated with body weight loss and decreased enjoyment of food. Besides taste, olfaction plays a role in food intake. We assessed whether systemic chemotherapeutic cancer treatment compromises olfactory function in rats and mice treated with docetaxel (Taxotere; Sanofi-Aventis, Paris, France). STUDY DESIGN: Randomized, controlled trials on mice and rats. METHODS: Male mice received a single and male rats either a single, two, or three docetaxel administrations. Olfactory function was tested by means of electroolfactograms (EOGs) from the chemosensory epithelium of the nasal septum and the endoturbinates. We evaluated and compared the magnitude of EOG responses evoked by different odorants recorded at different time points after treatment. RESULTS: In both animal species, docetaxel administration reduced body weight gain, thus evidencing the general toxic effect of the drug. In both animal species, the olfactory mucosa remained responsive to stimulation of odorants during the whole course of experiment, but treatment revealed regional differences of docetaxel susceptibility and induced marked transitory electrophysiological changes. In mice and rats a significant transitory decrease in EOG response magnitude occurred after a single administration. Unexpectedly, in rats we also observed an increase of the olfactory response following the second administration of the drug. CONCLUSIONS: Docetaxel exerts a neurotoxic effect on olfactory epithelia of rodents at doses similar to human doses, thus inducing transitory functional alterations. Although moderate, they are consistent with the hypothesis of a dysfunction of olfactory function. Further experiments are needed to elucidate the origin of the electrophysiological effects and their impact on the olfactory perception.

Comparing the information content of two large olfactory databases.

The expert's subjectivity in establishing an olfactory description can produce wide discrepancies in different databases listing the odor profile of identical compounds. A representative example is obtained by comparing the odorous compounds included in the "Perfumery Materials and Performance 2001" (PMP2001) database and in Arctander's books (1960 and 1969). To better assess this problem, classification models obtained by using the adaptive fuzzy partition method were established on subsets of these databases distributed into the same olfactory classes. The robustness and the prediction power of these models give a powerful criterion for evaluating the "quality" of their information content and for deciding which is the most trustable database.

Newborn rabbit responsiveness to the mammary pheromone is concentration-dependent.

The effect of the intensity of odour signals has rarely been investigated in the regulation of odour-guided behaviour in young mammals. This series of experiments used the mammary pheromone (MP) of the female rabbit to assess the influence of stimulus concentration on neonatal pup responsiveness. The MP is a single compound isolated from rabbit milk that releases in pups the typical head searching and oral seizing behaviour. The pups (n = 621) were exposed to graded concentrations of the MP in bioassays varying in stimulus delivery conditions. Experiment 1 demonstrated that in aqueous dilutions the MP efficiently elicits behavioural responses only within a limited range of concentrations (from 2.5 x 10(-9) to 2.5 x 10(-5) g/ml). Experiment 2 yielded the same outcome with highly purified MP delivered in dynamic conditions with a gas chromatograph. Finally, Experiment 3 used deodorized milk as the solvent of the MP; despite this change in the physico-chemical context of stimulation, similar results were reached.

Analysis of the functional maturation of olfactory neurons in chicks before and after birth.

There has been indirect evidence that the olfactory system of mammals could be functional shortly before birth. Taking advantage of the accessibility of bird embryos, we studied the functional maturation of the olfactory mucosa during embryonic development in birds. Using the combination of electrophysiological EOG recordings and immunohistochemical studies, it was possible to directly demonstrate for the first time that the olfactory system is functional during embryogenesis from embryonic day (ED) 13 and that the beginning of olfactory function coincides with the first localization of the calcium dependent calmodulin kinase II (CaMKIIalpha) in the dendrites of the olfactory receptor neurons. CaMKII and olfactory receptor genes are expressed much earlier in olfactory neurons, both involved in the sensory transduction, but the pattern of expression of CaMKIIalpha changes during the ontogenesis. The increase of EOG amplitude between ED13 and ED15 also coincides with the increase of the number of neurons presenting the dendritic localization of CaMKIIalpha. These results suggest that the enzyme CaMKII might play a role in the functional maturation of the olfactory mucosa.

Rhythm sequence through the olfactory bulb layers during the time window of a respiratory cycle.

The mammalian olfactory bulb is characterized by prominent oscillatory activity of its local field potentials. Breathing imposes the most important rhythm. Other rhythms have been described in the beta- and gamma-frequency ranges. We recorded unitary activities in different bulbar layers simultaneously with local field potentials in order to examine the different relationships existing between (i) breathing and field potential oscillations, and (ii) breathing and spiking activity of different cell types. We show that, whatever the layer, odour-induced gamma oscillations always occur around the transition point between inhalation and exhalation while beta oscillations appear during early exhalation and may extend up to the end of inhalation. By contrast, unitary activities exhibit different characteristics according to the layer. They vary in (i) their temporal relationship with respect to the respiratory cycle; (ii) their spike rates; (iii) their temporal patterns defined according to the respiratory cycle. The time window of a respiratory cycle might thus be split into three main epochs based on the deceleration of field potential rhythms (from gamma to beta oscillations) and a simultaneous gradient of spike discharge frequencies ranging from 180 to 30 Hz. We discuss the possibility that each rhythm could serve different functions as priming, gating or tuning for the bulbar network.